正在加载视频...

视频加载失败

返回首页

Been designing and experimenting with a new benchmark that stresses an underexplored angle: long tool-call chains with traps. The task: audit 36 packets, read 4 long-context ledgers, dodge retired/staging/wrong-quarter decoys, follow a strict workflow (auth → token → request → answer), submit the exact secret. Optimal: 52 calls. No...

stevibe's profile picture

stevibe

27,270 subscribers

18,438 次观看 • 3 个月前 •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Which local models can actually handle tool calling? I built a framework to find out. 15 scenarios. 12 tools. Mocked responses. Temperature 0. No cherry-picking. Tested every Qwen3.5 size from 0.8B to 397B, and since some of you asked after the distillation tests: yes, I included Jackrong's Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled too. Only two models went all green: the 27B dense and the distilled 27B. The 397B? Failed two tests. The 122B? Failed one. The 35B? Failed two. The timed-out results — mostly on the smaller models, are cases where the model got stuck in a loop, repeating the same tool call until it hit the 30-second limit. The test that exposed the most models: "Search for Iceland's population, then calculate 2% of it." Simple, but 35B, 122B, and 397B all used a rounded number from memory instead of the actual search result. They didn't trust their own tool output. Small models hallucinate data. Big models ignore data. The 27B just threaded it through.
0:11

Which local models can actually handle tool calling? I built a framework to find out. 15 scenarios. 12 tools. Mocked responses. Temperature 0. No cherry-picking. Tested every Qwen3.5 size from 0.8B to 397B, and since some of you asked after the distillation tests: yes, I included Jackrong's Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled too. Only two models went all green: the 27B dense and the distilled 27B. The 397B? Failed two tests. The 122B? Failed one. The 35B? Failed two. The timed-out results — mostly on the smaller models, are cases where the model got stuck in a loop, repeating the same tool call until it hit the 30-second limit. The test that exposed the most models: "Search for Iceland's population, then calculate 2% of it." Simple, but 35B, 122B, and 397B all used a rounded number from memory instead of the actual search result. They didn't trust their own tool output. Small models hallucinate data. Big models ignore data. The 27B just threaded it through.

stevibe

432,712 次观看 • 4 个月前

Qwen3.5 27B vs Gemma4 31B | Canvas Creativity Test Why HTML Canvas? Two reasons: 1. It's unforgiving, one small mistake and the whole thing breaks 2. We kept prompts short to test real creativity, not instruction following 4 rounds: - Analog Clock - Hyperspace Tunnel - Growing Tree - Black Hole Both nailed the clock, but the other three is where it gets interesting. Looking forward to Qwen3.6 open-weight release!
0:17

Qwen3.5 27B vs Gemma4 31B | Canvas Creativity Test Why HTML Canvas? Two reasons: 1. It's unforgiving, one small mistake and the whole thing breaks 2. We kept prompts short to test real creativity, not instruction following 4 rounds: - Analog Clock - Hyperspace Tunnel - Growing Tree - Black Hole Both nailed the clock, but the other three is where it gets interesting. Looking forward to Qwen3.6 open-weight release!

stevibe

170,768 次观看 • 4 个月前

Qwen3.6 35B A3B can't fill out a paper form on its own. But give it NVIDIA's LocateAnything-3B — the #1 trending model on HuggingFace — as its eyes, and the two small models get it done together. (The test: place each element at the right pixel position on a blank form image, not type into a field.) Setup: > Qwen is the brain (main model), LocateAnything is the eyes (helper model acting as a tool). > I gave Qwen a new tool: ask "where's the email field?" and LocateAnything returns the exact x, y, width, height. > The blue boxes on the screen are its detections. Look how tight they are — it nails every field. Result: > Qwen3.6 35B A3B + LocateAnything-3B: form completed, all info correct. > Name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code: all landed in the right field areas. > Character-box alignment still a touch loose, but every value is where it belongs. > 9m10s, 224.5k input, 24.3k output, 21 turns. Why it matters: > Qwen alone can't finish this test. Bolt on a 3B model that does exactly one thing > locate > and suddenly it can. > A combination of small models can do the work of a single large one.
0:14

Qwen3.6 35B A3B can't fill out a paper form on its own. But give it NVIDIA's LocateAnything-3B — the #1 trending model on HuggingFace — as its eyes, and the two small models get it done together. (The test: place each element at the right pixel position on a blank form image, not type into a field.) Setup: > Qwen is the brain (main model), LocateAnything is the eyes (helper model acting as a tool). > I gave Qwen a new tool: ask "where's the email field?" and LocateAnything returns the exact x, y, width, height. > The blue boxes on the screen are its detections. Look how tight they are — it nails every field. Result: > Qwen3.6 35B A3B + LocateAnything-3B: form completed, all info correct. > Name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code: all landed in the right field areas. > Character-box alignment still a touch loose, but every value is where it belongs. > 9m10s, 224.5k input, 24.3k output, 21 turns. Why it matters: > Qwen alone can't finish this test. Bolt on a 3B model that does exactly one thing > locate > and suddenly it can. > A combination of small models can do the work of a single large one.

stevibe

150,173 次观看 • 2 个月前

I explored a further possibility with local models: Qwen3.6 35B A3B + NVIDIA LocateAnything-3B as a local Computer Use agent (proof of concept). In the demo, I asked it to switch my Mac to light mode. It did. Then back to dark. Did that too — finding the right toggle in System Settings, clicking it, and verifying the change itself. It's fully screenshot-based, so no Accessibility API needed. If it's on screen, the agent can see it and act on it. This runs entirely on your own hardware — private, local, built from two small open models.
0:16

I explored a further possibility with local models: Qwen3.6 35B A3B + NVIDIA LocateAnything-3B as a local Computer Use agent (proof of concept). In the demo, I asked it to switch my Mac to light mode. It did. Then back to dark. Did that too — finding the right toggle in System Settings, clicking it, and verifying the change itself. It's fully screenshot-based, so no Accessibility API needed. If it's on screen, the agent can see it and act on it. This runs entirely on your own hardware — private, local, built from two small open models.

stevibe

44,116 次观看 • 2 个月前

my 8 GB VRAM gaming laptop is absolutely going to hate me for this. but I still did it. ran a 31b dense model (Gemma 4 31b Q4) with only 8 GB VRAM last week I ran Gemma 4 26B A4B a mixture of experts model on my RTX 4060 and hit 25–28 tokens/sec using llama.cpp's new MTP support. smooth. snappy. but MoE has a secret: it only activates 4B parameters per token despite having 26B total. that's why it flies. so the real question started haunting me. what if I throw a full, no tricks, every parameter fires on every token, 31B DENSE model at the same machine? # Hardware: GPU: NVIDIA RTX 4060, 8 GB VRAM RAM: 16 GB CPU: Intel Core i7 H Laptop. Gaming. Modest. The model: gemma-4-31B-it-qat-UD-Q4_K_XL.gguf (model's unsloth huggingface link in the comments) This is Google DeepMind's flagship dense model in the Gemma 4 family that can run on single consumer GPU. It packs a hybrid attention architecture, supports up to 256K context natively, and is QAT (Quantization Aware Training) optimized, meaning it retains far more quality than standard post training quants at the same bit depth. This is NOT the MoE. This is 31 BILLION dense parameters, every single one of them loaded. # the flags I used: -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -cnv --spec-type draft-mtp --spec-draft-model mtp-gemma-4-31B-it.gguf --spec-draft-n-max 8 --spec-draft-p-min 0.6 -c 6000 -v Multi Token Prediction (MTP) is still active here. Separate draft GGUF required, same as the 26B setup. # Results: → Decode: ~3 tokens/sec → Prefill: ~2 tokens/sec → Context: 6000 tokens → Hardware crying quietly in the corner: yes so is 3 tps actually usable? For real time back and forth chat? Not ideal. You're not having a fluid conversation at 3 tps. but slow ≠ useless. And this is where it gets genuinely interesting. think about how senior devs actually work in a real team. But when something is architectural, deeply complex, or needs serious reasoning? they walk down the hall and escalate to the senior. That's exactly the local AI agent architecture this unlocks: → Fast orchestrator model (Gemma 4 26B MoE at 25+ tps) handles routing, simple queries, tool calls, memory. The junior dev. → Gemma 4 31B dense is the senior, called only when the fast model genuinely hits a wall. Hard multi step reasoning. Complex code generation. Deep architectural decisions. The agentic loop stays fast. Only the hard hops touch the 31B. That's a legitimate production grade local AI architecture on a budget hardware. (requires 2 8gb gpus) other workflows where 3 tps is completely fine: - overnight batch jobs. summarize documents, extract structured data, review code. Fire it off. Sleep. wake up to results. - One shot deep reasoning - Silent code audit loops, you write and test, the 31B reviews diffs and flags issues in the background between your sprints - Any workflow where output quality > output speed A few weeks ago, nobody was running a 30B+ dense model on a single consumer GPU with 8 GB VRAM. At all. Now we're doing it on an Intel i7-H gaming laptop with a NVIDIA RTX 4060, thanks to llama.cpp + QAT quants + MTP speculative drafting. Google DeepMind said the Gemma 4 31B targets "consumer GPUs and workstations." They were not exaggerating. The hardware bar to run serious frontier class models locally keeps dropping. the tools are here. the models are here. you just have to be willing to abuse your laptop a little. what workflows would you actually run on a local 3 tps 31B dense model? genuinely curious. drop it below.
0:18

my 8 GB VRAM gaming laptop is absolutely going to hate me for this. but I still did it. ran a 31b dense model (Gemma 4 31b Q4) with only 8 GB VRAM last week I ran Gemma 4 26B A4B a mixture of experts model on my RTX 4060 and hit 25–28 tokens/sec using llama.cpp's new MTP support. smooth. snappy. but MoE has a secret: it only activates 4B parameters per token despite having 26B total. that's why it flies. so the real question started haunting me. what if I throw a full, no tricks, every parameter fires on every token, 31B DENSE model at the same machine? # Hardware: GPU: NVIDIA RTX 4060, 8 GB VRAM RAM: 16 GB CPU: Intel Core i7 H Laptop. Gaming. Modest. The model: gemma-4-31B-it-qat-UD-Q4_K_XL.gguf (model's unsloth huggingface link in the comments) This is Google DeepMind's flagship dense model in the Gemma 4 family that can run on single consumer GPU. It packs a hybrid attention architecture, supports up to 256K context natively, and is QAT (Quantization Aware Training) optimized, meaning it retains far more quality than standard post training quants at the same bit depth. This is NOT the MoE. This is 31 BILLION dense parameters, every single one of them loaded. # the flags I used: -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -cnv --spec-type draft-mtp --spec-draft-model mtp-gemma-4-31B-it.gguf --spec-draft-n-max 8 --spec-draft-p-min 0.6 -c 6000 -v Multi Token Prediction (MTP) is still active here. Separate draft GGUF required, same as the 26B setup. # Results: → Decode: ~3 tokens/sec → Prefill: ~2 tokens/sec → Context: 6000 tokens → Hardware crying quietly in the corner: yes so is 3 tps actually usable? For real time back and forth chat? Not ideal. You're not having a fluid conversation at 3 tps. but slow ≠ useless. And this is where it gets genuinely interesting. think about how senior devs actually work in a real team. But when something is architectural, deeply complex, or needs serious reasoning? they walk down the hall and escalate to the senior. That's exactly the local AI agent architecture this unlocks: → Fast orchestrator model (Gemma 4 26B MoE at 25+ tps) handles routing, simple queries, tool calls, memory. The junior dev. → Gemma 4 31B dense is the senior, called only when the fast model genuinely hits a wall. Hard multi step reasoning. Complex code generation. Deep architectural decisions. The agentic loop stays fast. Only the hard hops touch the 31B. That's a legitimate production grade local AI architecture on a budget hardware. (requires 2 8gb gpus) other workflows where 3 tps is completely fine: - overnight batch jobs. summarize documents, extract structured data, review code. Fire it off. Sleep. wake up to results. - One shot deep reasoning - Silent code audit loops, you write and test, the 31B reviews diffs and flags issues in the background between your sprints - Any workflow where output quality > output speed A few weeks ago, nobody was running a 30B+ dense model on a single consumer GPU with 8 GB VRAM. At all. Now we're doing it on an Intel i7-H gaming laptop with a NVIDIA RTX 4060, thanks to llama.cpp + QAT quants + MTP speculative drafting. Google DeepMind said the Gemma 4 31B targets "consumer GPUs and workstations." They were not exaggerating. The hardware bar to run serious frontier class models locally keeps dropping. the tools are here. the models are here. you just have to be willing to abuse your laptop a little. what workflows would you actually run on a local 3 tps 31B dense model? genuinely curious. drop it below.

Alok

63,689 次观看 • 2 个月前

How well can Qwen3.5 models debug code? I built BugFind-15 — 15 buggy snippets across Python, JS, Rust, and Go. Docker sandbox compiles and validates every fix. Two trap scenarios where the code is correct and the model must resist "fixing" it. Tested every Qwen3.5 size from 0.8B to 397B, plus Jackrong's popular distilled model (V2). The 0.8B scored 5%. The 2B scored 10%. At 4B, debugging ability jumps to 69%. The hardest scenario: BF-03, a Rust trap. The code compiles fine — format! borrows, it doesn't move. Not a single model figured this out. From 0.8B to 397B, every one of them "fixed" a bug that doesn't exist. Category C (subtle bugs — mutable defaults, integer overflow, slice aliasing) was 100% across every model 4B and above. Category D (red herring resistance) told the real story — can it resist fixing code that isn't broken? No model scored above 90%. Small models can't debug. Mid-size models fix obvious bugs but fall for traps. Large models fix the hard bugs but still invent problems that don't exist.
0:11

How well can Qwen3.5 models debug code? I built BugFind-15 — 15 buggy snippets across Python, JS, Rust, and Go. Docker sandbox compiles and validates every fix. Two trap scenarios where the code is correct and the model must resist "fixing" it. Tested every Qwen3.5 size from 0.8B to 397B, plus Jackrong's popular distilled model (V2). The 0.8B scored 5%. The 2B scored 10%. At 4B, debugging ability jumps to 69%. The hardest scenario: BF-03, a Rust trap. The code compiles fine — format! borrows, it doesn't move. Not a single model figured this out. From 0.8B to 397B, every one of them "fixed" a bug that doesn't exist. Category C (subtle bugs — mutable defaults, integer overflow, slice aliasing) was 100% across every model 4B and above. Category D (red herring resistance) told the real story — can it resist fixing code that isn't broken? No model scored above 90%. Small models can't debug. Mid-size models fix obvious bugs but fall for traps. Large models fix the hard bugs but still invent problems that don't exist.

stevibe

35,158 次观看 • 4 个月前

I went a little overboard with Codex last week and burned through my entire weekly allowance in two days. Luckily, my quota reset today. Otherwise, I’m not sure what I would’ve done. It got me thinking: instead of asking one large model to handle everything from start to finish, why not let a stronger model plan the project and review the work, while a model built for execution handles the day-to-day implementation? So I tried it. The result was better than I expected. I used GPT-5.6 Sol in Codex as the decision-maker, then ran Ling-3.0-flash from Ant Ling inside OpenCode as the execution engine. Together, they built a small 3D farming game. Before writing any code, I had Codex create four documents: SPEC.md defined the product scope and the lines we couldn’t cross. ARCHITECTURE.md laid out the isometric coordinate system, state machine, and module boundaries. TASKS.md broke the project into small jobs Ling could tackle one at a time. ACCEPTANCE.md explained how each step would be tested and what “done” actually meant. Then I gave Ling a very straightforward role: You are the execution model for this project. Read all four documents before you begin. Work only on the task assigned for this round. When you’re done, run typecheck, test, and build. If anything fails, read the error, fix it, and run the checks again. Do not move on to the next task early. Ling handled dependency installation, project structure, strict TypeScript configuration, test setup, and a production build in 6 minutes and 3 seconds. It ran into issues with the Vite test config, a TS6310 error, and a missing jsdom dependency along the way. Instead of stopping at the first error, it kept reading the logs and fixing the problems until all three checks passed. The speed was honestly hard to believe. If you exclude the time spent waiting on tools, it was producing more than 100 tokens per second. That made the whole development loop feel noticeably faster. After this experiment, I’m planning to keep using the same workflow. If the task is small, there’s no reason to call an expensive planning model for every single step. If the task is large, handing the entire project to a Flash model in one prompt isn’t a great idea either. The setup that makes more sense to me is: Use a more capable model such as Codex to explore the project, make architectural decisions, and break the work down. Put the constraints into specs, schemas, types, and tests instead of leaving them buried in chat history. Give Ling-3.0-flash a steady stream of clear, verifiable implementation tasks. Report bugs with structured context and actual error logs, rather than saying, “It still doesn’t work.” Bring Codex back in for architecture reviews, visual checks, and changes that affect multiple parts of the project. The point of this setup isn’t to give AI a big “build the whole project” button. It’s to turn software development into a pipeline with a much more sensible cost structure: Codex figures out the plan, sets the boundaries, and catches problems. Ling-3.0-flash moves quickly, calls tools reliably, and works through well-defined tasks at scale. For agent workflows that involve lots of repetitive edits, production tasks, and tool calls, this may be a more practical answer than simply using the biggest model for everything.
0:16

I went a little overboard with Codex last week and burned through my entire weekly allowance in two days. Luckily, my quota reset today. Otherwise, I’m not sure what I would’ve done. It got me thinking: instead of asking one large model to handle everything from start to finish, why not let a stronger model plan the project and review the work, while a model built for execution handles the day-to-day implementation? So I tried it. The result was better than I expected. I used GPT-5.6 Sol in Codex as the decision-maker, then ran Ling-3.0-flash from Ant Ling inside OpenCode as the execution engine. Together, they built a small 3D farming game. Before writing any code, I had Codex create four documents: SPEC.md defined the product scope and the lines we couldn’t cross. ARCHITECTURE.md laid out the isometric coordinate system, state machine, and module boundaries. TASKS.md broke the project into small jobs Ling could tackle one at a time. ACCEPTANCE.md explained how each step would be tested and what “done” actually meant. Then I gave Ling a very straightforward role: You are the execution model for this project. Read all four documents before you begin. Work only on the task assigned for this round. When you’re done, run typecheck, test, and build. If anything fails, read the error, fix it, and run the checks again. Do not move on to the next task early. Ling handled dependency installation, project structure, strict TypeScript configuration, test setup, and a production build in 6 minutes and 3 seconds. It ran into issues with the Vite test config, a TS6310 error, and a missing jsdom dependency along the way. Instead of stopping at the first error, it kept reading the logs and fixing the problems until all three checks passed. The speed was honestly hard to believe. If you exclude the time spent waiting on tools, it was producing more than 100 tokens per second. That made the whole development loop feel noticeably faster. After this experiment, I’m planning to keep using the same workflow. If the task is small, there’s no reason to call an expensive planning model for every single step. If the task is large, handing the entire project to a Flash model in one prompt isn’t a great idea either. The setup that makes more sense to me is: Use a more capable model such as Codex to explore the project, make architectural decisions, and break the work down. Put the constraints into specs, schemas, types, and tests instead of leaving them buried in chat history. Give Ling-3.0-flash a steady stream of clear, verifiable implementation tasks. Report bugs with structured context and actual error logs, rather than saying, “It still doesn’t work.” Bring Codex back in for architecture reviews, visual checks, and changes that affect multiple parts of the project. The point of this setup isn’t to give AI a big “build the whole project” button. It’s to turn software development into a pipeline with a much more sensible cost structure: Codex figures out the plan, sets the boundaries, and catches problems. Ling-3.0-flash moves quickly, calls tools reliably, and works through well-defined tasks at scale. For agent workflows that involve lots of repetitive edits, production tasks, and tool calls, this may be a more practical answer than simply using the biggest model for everything.

雪踏乌云

23,107 次观看 • 22 天前

Introducing HermesAgent-20, a new Bench Pack for BenchLocal. 20 scenarios extracted straight from the Hermes Agent source code, run against a REAL Hermes instance. The actual workload you'd put your model through. Why I built BenchLocal in the first place: most benchmarks are too abstract. We use local LLMs for practical work, and finding the right model for YOUR task efficiently is the single most important thing, especially when you're constrained to what fits on your machine. BenchLocal is a framework: providers, models, side-by-side comparison, all in one UI. Bench Packs are the unit of testing: ToolCall-15 and BugFind-15 shipped first, and when I launched the BenchLocal 0.1.0, added StructOutput, ReasonMath, InstructFollow, DataExtract. Now, HermesAgent-20 is the newest. Bench Packs install like VS Code extensions. The SDK is open, write your own, share it, grow the ecosystem. Here's the goal: a community-built, practical evaluation layer for the local LLM space. Early numbers on HermesAgent-20: > GLM 5.1 — 85 > Gemma4 31B — 83 > Qwen3.5 27B — 79 > MiniMax M2.7 — 76 Upgrade to the latest BenchLocal to install HermesAgent-20 (SDK update required).
0:20

Introducing HermesAgent-20, a new Bench Pack for BenchLocal. 20 scenarios extracted straight from the Hermes Agent source code, run against a REAL Hermes instance. The actual workload you'd put your model through. Why I built BenchLocal in the first place: most benchmarks are too abstract. We use local LLMs for practical work, and finding the right model for YOUR task efficiently is the single most important thing, especially when you're constrained to what fits on your machine. BenchLocal is a framework: providers, models, side-by-side comparison, all in one UI. Bench Packs are the unit of testing: ToolCall-15 and BugFind-15 shipped first, and when I launched the BenchLocal 0.1.0, added StructOutput, ReasonMath, InstructFollow, DataExtract. Now, HermesAgent-20 is the newest. Bench Packs install like VS Code extensions. The SDK is open, write your own, share it, grow the ecosystem. Here's the goal: a community-built, practical evaluation layer for the local LLM space. Early numbers on HermesAgent-20: > GLM 5.1 — 85 > Gemma4 31B — 83 > Qwen3.5 27B — 79 > MiniMax M2.7 — 76 Upgrade to the latest BenchLocal to install HermesAgent-20 (SDK update required).

stevibe

38,631 次观看 • 4 个月前

I just got Gemma 4 26B A4B MoE model running fully locally with Hermes agent on an 8GB RTX 4060 and it's now backtesting trading strategies end to end, no hand holding. If you’re a trader or work on Wall Street, you don’t want to miss this. Yes. fully automated. No cloud. No APIs beyond market data. # Here's what I did: Setup: - Model: Gemma 4 26B-A4B QAT (MoE), Q4_K_XL Unsloth's quant (link in the comments) - Inference: llama.cpp (turboquant fork by Tom Turney link in the comments) - Hardware: RTX 4060, 8GB VRAM + 16GB RAM only (with 50 other chrome tabs open) - Context: 64K llama.cpp turboquant flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 64000 --cache-type-k q8_0 --cache-type-v turbo3 --port 8080 turboquant helps achieve high prefill and decode throughput for interactive sessions. throughput with Hermes agent: decode: 25+ tokens/sec prefill: 250+ tokens/sec # Then I gave the agent one task: Backtest a strategy: - Buy when RSI crosses above 30 - Sell at +2% profit or -1% stoploss - No overlapping positions - Use Google stock via yfinance - Generate a full HTML report with candlestick charts + signals What happened next was wild. It didn't just write code, it ran the entire workflow itself: Audited the environment (pip list, dependency check) Hit a ModuleNotFoundError, multiple Python installs were conflicting Ran where python to map every interpreter on the system Manually selected the correct Python 3.13 path and re ran the script Wrote a clean statevmachine backtester (strict no overlapping trades logic) Patched a yfinance MultiIndex quirk that would've crashed the script Built Plotly candlestick + RSI charts with buy/sell markers Calculated win rate, PnL, and summary stats Exported a polished single file HTML report. check the report at the end of the video or in the comments. Biggest takeaway: local LLMs aren't just "chat assistants" anymore. They debug their own environment, write production code, and ship a finished deliverable on consumer hardware, for $0 in API costs. If you're still calling local models "toys," you're already behind. This is just the beginning. Hermes agent just surpassed 1 trillion tokens in a single day on OpenRouter. Think about the scale of total token generation happening right now. Disclaimer: This is not financial advice. Consult a professional before making any trading decisions.
0:14

I just got Gemma 4 26B A4B MoE model running fully locally with Hermes agent on an 8GB RTX 4060 and it's now backtesting trading strategies end to end, no hand holding. If you’re a trader or work on Wall Street, you don’t want to miss this. Yes. fully automated. No cloud. No APIs beyond market data. # Here's what I did: Setup: - Model: Gemma 4 26B-A4B QAT (MoE), Q4_K_XL Unsloth's quant (link in the comments) - Inference: llama.cpp (turboquant fork by Tom Turney link in the comments) - Hardware: RTX 4060, 8GB VRAM + 16GB RAM only (with 50 other chrome tabs open) - Context: 64K llama.cpp turboquant flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 64000 --cache-type-k q8_0 --cache-type-v turbo3 --port 8080 turboquant helps achieve high prefill and decode throughput for interactive sessions. throughput with Hermes agent: decode: 25+ tokens/sec prefill: 250+ tokens/sec # Then I gave the agent one task: Backtest a strategy: - Buy when RSI crosses above 30 - Sell at +2% profit or -1% stoploss - No overlapping positions - Use Google stock via yfinance - Generate a full HTML report with candlestick charts + signals What happened next was wild. It didn't just write code, it ran the entire workflow itself: Audited the environment (pip list, dependency check) Hit a ModuleNotFoundError, multiple Python installs were conflicting Ran where python to map every interpreter on the system Manually selected the correct Python 3.13 path and re ran the script Wrote a clean statevmachine backtester (strict no overlapping trades logic) Patched a yfinance MultiIndex quirk that would've crashed the script Built Plotly candlestick + RSI charts with buy/sell markers Calculated win rate, PnL, and summary stats Exported a polished single file HTML report. check the report at the end of the video or in the comments. Biggest takeaway: local LLMs aren't just "chat assistants" anymore. They debug their own environment, write production code, and ship a finished deliverable on consumer hardware, for $0 in API costs. If you're still calling local models "toys," you're already behind. This is just the beginning. Hermes agent just surpassed 1 trillion tokens in a single day on OpenRouter. Think about the scale of total token generation happening right now. Disclaimer: This is not financial advice. Consult a professional before making any trading decisions.

Alok

105,094 次观看 • 1 个月前

GLM 5.1 just went open-weight on Hugging Face, but how does it compare to GLM 5? I have tested both with the canvas tree challenge. 5.1 thinks longer, but delivers wind animation, sun, clouds, and way more detail. Prompt attached: Write a single HTML file with a full-page canvas, no libraries. Animate a tree that grows from the bottom center of the screen in real time. The trunk grows upward first, then branches split off recursively with slight randomness in angle and length. Each generation of branches should be thinner and slightly lighter in color. When branches reach their final size, add small leaves as soft green circles at the tips. The tree should take about 15 seconds to fully grow. Use warm brown for wood and varied greens for leaves against a soft sky-blue gradient background.
0:19

GLM 5.1 just went open-weight on Hugging Face, but how does it compare to GLM 5? I have tested both with the canvas tree challenge. 5.1 thinks longer, but delivers wind animation, sun, clouds, and way more detail. Prompt attached: Write a single HTML file with a full-page canvas, no libraries. Animate a tree that grows from the bottom center of the screen in real time. The trunk grows upward first, then branches split off recursively with slight randomness in angle and length. Each generation of branches should be thinner and slightly lighter in color. When branches reach their final size, add small leaves as soft green circles at the tips. The tree should take about 15 seconds to fully grow. Use warm brown for wood and varied greens for leaves against a soft sky-blue gradient background.

stevibe

46,657 次观看 • 4 个月前

Don't train the model, evolve the harness. I read a brilliant blog post from Hugging Face where they took a frozen open model scoring 0% on a hard legal agent benchmark, left its weights alone, and let an automated loop rewrite only the code around it. That code layer is the harness, the runtime wrapper that feeds the model context, runs its tool calls, and decides when a run ends. By the time the loop finished, the system had essentially matched Sonnet 4.6 on the benchmark's headline metric, at roughly 7x lower cost per task. Zero weights changed. The gain existed because of where the model was failing. The judge only grades files saved in the right place under the exact requested filename, and the model kept doing the legal analysis correctly, then saving it under the wrong name, dropping it in a scratch folder, or never writing it at all. So the 0% was never measuring legal reasoning. It was measuring the harness. Hand-tuning that layer is slow and model-specific, so they automated it. A Claude proposer adds exactly one mechanism per iteration, and an outer loop keeps it only if it clearly beats the current best, so accepted mechanisms compound. What the loop discovered says a lot about where agents actually fail. → The biggest single gain was file handling, not intelligence. An automatic step that lands the deliverable exactly where the judge expects it beat every prompt change, with zero extra model tokens. → Code fixes transferred across models, prompt playbooks did not. The same harness lifted a smaller model from the same family by 14 points, but the tuned prompts hurt a different model family on tasks it could already finish. → The harness mattered more than anything else. Same model, same judge, same tasks, and five different harnesses scored anywhere between 3.5% and 80.1%. The gains do eventually flatten, and the remaining misses look like real capability gaps. At some point the wrapper runs out of tricks and the model has to carry the work. But the lesson holds. A benchmark score measures the model and its harness together, and until the harness is fixed, it's impossible to know which one failed. I highly recommend reading this: I also wrote a deep dive on agent harness engineering a while back, covering the orchestration loop, tools, memory, context management, and everything that turns a stateless LLM into a capable agent. The article is quoted below.
0:18

Don't train the model, evolve the harness. I read a brilliant blog post from Hugging Face where they took a frozen open model scoring 0% on a hard legal agent benchmark, left its weights alone, and let an automated loop rewrite only the code around it. That code layer is the harness, the runtime wrapper that feeds the model context, runs its tool calls, and decides when a run ends. By the time the loop finished, the system had essentially matched Sonnet 4.6 on the benchmark's headline metric, at roughly 7x lower cost per task. Zero weights changed. The gain existed because of where the model was failing. The judge only grades files saved in the right place under the exact requested filename, and the model kept doing the legal analysis correctly, then saving it under the wrong name, dropping it in a scratch folder, or never writing it at all. So the 0% was never measuring legal reasoning. It was measuring the harness. Hand-tuning that layer is slow and model-specific, so they automated it. A Claude proposer adds exactly one mechanism per iteration, and an outer loop keeps it only if it clearly beats the current best, so accepted mechanisms compound. What the loop discovered says a lot about where agents actually fail. → The biggest single gain was file handling, not intelligence. An automatic step that lands the deliverable exactly where the judge expects it beat every prompt change, with zero extra model tokens. → Code fixes transferred across models, prompt playbooks did not. The same harness lifted a smaller model from the same family by 14 points, but the tuned prompts hurt a different model family on tasks it could already finish. → The harness mattered more than anything else. Same model, same judge, same tasks, and five different harnesses scored anywhere between 3.5% and 80.1%. The gains do eventually flatten, and the remaining misses look like real capability gaps. At some point the wrapper runs out of tricks and the model has to carry the work. But the lesson holds. A benchmark score measures the model and its harness together, and until the harness is fixed, it's impossible to know which one failed. I highly recommend reading this: I also wrote a deep dive on agent harness engineering a while back, covering the orchestration loop, tools, memory, context management, and everything that turns a stateless LLM into a capable agent. The article is quoted below.

Akshay 🚀

244,567 次观看 • 1 个月前

you're paying $20/mo for something your $500 GPU can already do. Gemma 4 26B A4B QAT MoE + Hermes Agent running on a single RTX 4060 (8GB VRAM). Built a vision capable, 100% free, 100% local, private AI assistant that lives in my Chrome browser. No API keys. No cloud. No subscriptions. 100% vibe coded. 0% handholding. It has full context of whatever's on my screen can answer questions, summarize pages, extract data, and see images. Same local model handles everything, no external calls, ever. keep reading for the model and hermes agent tips i learnt while building this locally. Here's the exact setup for anyone running local LLMs on 6-8 GB VRAM: llama.cpp server flags (on my NVIDIA RTX 4060 8gb VRAM): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --cache-type-k q8_0 --cache-type-v q8_0 -c 150000 --port 8080 Throughput with quantization: Prefill: 200-250 tokens/sec Decode: 20-25 tokens/sec reduce context if oom on 6 gb vram card. Key learnings: - Quantize KV cache to q8 for faster prefill/decode. Prefill goes from 100-150 (unquantized) to 200-250 tok/s (q8). - But watch out, once actual context grows past ~50k tokens on high entropy workloads, q8 KV quantization can cause hallucinations. Low entropy workloads are mostly unaffected. If you see it happening, drop the quantization. This is common across all local models. - In Hermes Agent settings -> Memory & Context, bump compression threshold from default 0.5 to 0.7. Default triggers way too frequent context compression and eats time. Up next: add persistent memory, web search, tool calling, streaming output and whatever you suggest. Running a 26B MoE with vision + 150k context window on 8GB VRAM would've sounded impossible 6 months ago. Works the same on the NVIDIA RTX 3060 Ti, 3070, 4060 Ti, 5060, 2080, or any 8GB card. VRAM is the only requirement. Local AI agents are closer than people think. You just need to know where the knobs are. Model's Unsloth quant hugging face link in the comments. Have you tried Hermes agent by Nous Research yet? What are you building with local LLMs? Drop it below, let's see what this community is shipping.
0:19

you're paying $20/mo for something your $500 GPU can already do. Gemma 4 26B A4B QAT MoE + Hermes Agent running on a single RTX 4060 (8GB VRAM). Built a vision capable, 100% free, 100% local, private AI assistant that lives in my Chrome browser. No API keys. No cloud. No subscriptions. 100% vibe coded. 0% handholding. It has full context of whatever's on my screen can answer questions, summarize pages, extract data, and see images. Same local model handles everything, no external calls, ever. keep reading for the model and hermes agent tips i learnt while building this locally. Here's the exact setup for anyone running local LLMs on 6-8 GB VRAM: llama.cpp server flags (on my NVIDIA RTX 4060 8gb VRAM): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --cache-type-k q8_0 --cache-type-v q8_0 -c 150000 --port 8080 Throughput with quantization: Prefill: 200-250 tokens/sec Decode: 20-25 tokens/sec reduce context if oom on 6 gb vram card. Key learnings: - Quantize KV cache to q8 for faster prefill/decode. Prefill goes from 100-150 (unquantized) to 200-250 tok/s (q8). - But watch out, once actual context grows past ~50k tokens on high entropy workloads, q8 KV quantization can cause hallucinations. Low entropy workloads are mostly unaffected. If you see it happening, drop the quantization. This is common across all local models. - In Hermes Agent settings -> Memory & Context, bump compression threshold from default 0.5 to 0.7. Default triggers way too frequent context compression and eats time. Up next: add persistent memory, web search, tool calling, streaming output and whatever you suggest. Running a 26B MoE with vision + 150k context window on 8GB VRAM would've sounded impossible 6 months ago. Works the same on the NVIDIA RTX 3060 Ti, 3070, 4060 Ti, 5060, 2080, or any 8GB card. VRAM is the only requirement. Local AI agents are closer than people think. You just need to know where the knobs are. Model's Unsloth quant hugging face link in the comments. Have you tried Hermes agent by Nous Research yet? What are you building with local LLMs? Drop it below, let's see what this community is shipping.

Alok

36,031 次观看 • 1 个月前

MiniMax M3 just dropped — their first natively multimodal model. So I ran it through my form-filling test. (The model has to place each element at the right pixel position on a blank form image, not type into a field.) Verdict: it got everything on the paper. > Name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code, all there. > Best character spacing I've seen yet: it actually calculates the gap between each character, clean across the DOB and number boxes > A few fields slightly misaligned, but every piece of data made it onto the form The reasoning chain is the interesting part: it does the easy fields first, then works into the tight one-char-per-box fields, reasoning through y-coordinates, baselines, and label clearance in obsessive detail. The cost: 40:33 and 126.7k output tokens. That's a long think — but it's MiniMax's first multimodal model, and it nailed the content.
0:14

MiniMax M3 just dropped — their first natively multimodal model. So I ran it through my form-filling test. (The model has to place each element at the right pixel position on a blank form image, not type into a field.) Verdict: it got everything on the paper. > Name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code, all there. > Best character spacing I've seen yet: it actually calculates the gap between each character, clean across the DOB and number boxes > A few fields slightly misaligned, but every piece of data made it onto the form The reasoning chain is the interesting part: it does the easy fields first, then works into the tight one-char-per-box fields, reasoning through y-coordinates, baselines, and label clearance in obsessive detail. The cost: 40:33 and 126.7k output tokens. That's a long think — but it's MiniMax's first multimodal model, and it nailed the content.

stevibe

27,383 次观看 • 2 个月前

MiniMax M3 might be the most underrated coding model right now. I gave it nothing but a screenshot of a chaotic 90s GeoCities-style fan page, no HTML source, just the image + the asset files, and told it to rebuild the whole thing as a sleek Apple-style 2026 site. One shot. Through OpenCode. The result is genuinely stunning. It kept the soul (the "stevibe's HyperHome" identity, the visitor counter, the guestbook, the webmaster portrait) and translated every section into clean modern design, gradient hero, proper typography, dark theme, the works.
0:19

MiniMax M3 might be the most underrated coding model right now. I gave it nothing but a screenshot of a chaotic 90s GeoCities-style fan page, no HTML source, just the image + the asset files, and told it to rebuild the whole thing as a sleek Apple-style 2026 site. One shot. Through OpenCode. The result is genuinely stunning. It kept the soul (the "stevibe's HyperHome" identity, the visitor counter, the guestbook, the webmaster portrait) and translated every section into clean modern design, gradient hero, proper typography, dark theme, the works.

stevibe

21,206 次观看 • 2 个月前

Got a 16GB GPU? You can run all of these right now. Tested 4 Qwen3.5-based models on ToolCall-15 & BugFind-15: Models: - Qwen3.5:9b Q8 (Official) - Qwopus v3 Q8 by Jackrong - OmniCoder-9B by Tesslate - Qwen3.5-9b-Sushi-Coder by bigatuna Summary: - ToolCall-15: Qwopus v3 went perfect 30/30, Sushicoder beat base Qwen3.5 - BugFind-15: Omnicoder flipped the script and took #1 at 83% No single model won both, that's the fun part. Open source community is cooking.
0:14

Got a 16GB GPU? You can run all of these right now. Tested 4 Qwen3.5-based models on ToolCall-15 & BugFind-15: Models: - Qwen3.5:9b Q8 (Official) - Qwopus v3 Q8 by Jackrong - OmniCoder-9B by Tesslate - Qwen3.5-9b-Sushi-Coder by bigatuna Summary: - ToolCall-15: Qwopus v3 went perfect 30/30, Sushicoder beat base Qwen3.5 - BugFind-15: Omnicoder flipped the script and took #1 at 83% No single model won both, that's the fun part. Open source community is cooking.

stevibe

75,514 次观看 • 4 个月前

Stanford researchers did it again. They just built the agent-native version of Git. When an agent works on a longer task, the run builds up a lot of state. This includes files edited/created, a dev server, a database, installed packages, KV cache, etc. Say the agent is at step 10 and makes a mistake, maybe it misreads a traceback and rewrites a file that was actually fine. The tests start failing, and the run goes off track, although everything through step eight was correct. By default, the agent just tries to fix it, which creates more edits and tool calls. This burns more tokens and grows the context. The other options are a person stepping in to redirect it or restarting the whole run from step one. That's wasteful, because it pays for every model/tool call again and re-prefills the context. Moreover, since an agent's run is non-deterministic, it doesn't reproduce the same early steps anyway. The reason it's hard to just jump back exactly to a previous correct step and resume from there is that the trajectory is only a message log. It records what the agent said and which tools it called, but not the live state underneath. That state includes things like memory, open file handles, child processes, installed packages, /tmp, and KV cache. None of that is in the log. Git can version the files, but it doesn't snapshot the running process or the KV cache. Checking out step eight moves the files back, but the process is still sitting in step-ten memory with a cold cache. Shepherd is a runtime layer by Stanford that records the run as a trace of typed events rather than a flat log. Each agent-environment interaction becomes a commit, similar to Git, but it tracks the live run. Its commit includes the agent process and the filesystem together, copy-on-write, so a branch carries the actual state and not just the files. Going back to a previous step is then a single call that forks from that commit and continues from the exact state. The copy-on-write fork is roughly five times faster than docker commit, and because the prompt prefix through step eight is unchanged, the KV cache is reused over 95% on replay, so early steps aren't reprocessed again. Once the run can be forked, a meta-agent can sit on top and operate it. It watches the trace and reverts as soon as it looks wrong, before the bad write is committed. In practice, it's just Python calling fork, replay, and revert on the trace, rather than a separate control plane wired into the harness. Not everything is reversible though. Files and sandbox changes undo themselves, but a database write has no automatic undo, so it needs a matching undo step set up in advance. Something external, like a sent email or a real charge, can't be undone, so the supervisor's job there is to catch it before it fires. They tested this on a few public benchmarks. On CooperBench, where two agents work on the same codebase, adding a live supervisor took the pair-coding pass rate from 28.8% to 54.7%. It's still early and labeled alpha. The benefit mostly shows up when a run gets branched a lot over a heavy sandbox state, which is exactly where restarting wastes the most tokens and time. If Git was made to make file changes reversible, Shepherd is trying to do the same thing for a live agent run. Shepherd Repo: (don't forget to star it ⭐ ) That said, Shepherd reverts a bad step inside a run. The harness around it, the prompts, tools, and checks the supervisor relies on, still drifts across runs as models and dependencies change. Akshay wrote about making that harness repair itself, where a failing trace gets diagnosed, the fix is verified against the exact input that failed, and the failure is locked as a regression test so it can't recur. Read it below.
0:15

Stanford researchers did it again. They just built the agent-native version of Git. When an agent works on a longer task, the run builds up a lot of state. This includes files edited/created, a dev server, a database, installed packages, KV cache, etc. Say the agent is at step 10 and makes a mistake, maybe it misreads a traceback and rewrites a file that was actually fine. The tests start failing, and the run goes off track, although everything through step eight was correct. By default, the agent just tries to fix it, which creates more edits and tool calls. This burns more tokens and grows the context. The other options are a person stepping in to redirect it or restarting the whole run from step one. That's wasteful, because it pays for every model/tool call again and re-prefills the context. Moreover, since an agent's run is non-deterministic, it doesn't reproduce the same early steps anyway. The reason it's hard to just jump back exactly to a previous correct step and resume from there is that the trajectory is only a message log. It records what the agent said and which tools it called, but not the live state underneath. That state includes things like memory, open file handles, child processes, installed packages, /tmp, and KV cache. None of that is in the log. Git can version the files, but it doesn't snapshot the running process or the KV cache. Checking out step eight moves the files back, but the process is still sitting in step-ten memory with a cold cache. Shepherd is a runtime layer by Stanford that records the run as a trace of typed events rather than a flat log. Each agent-environment interaction becomes a commit, similar to Git, but it tracks the live run. Its commit includes the agent process and the filesystem together, copy-on-write, so a branch carries the actual state and not just the files. Going back to a previous step is then a single call that forks from that commit and continues from the exact state. The copy-on-write fork is roughly five times faster than docker commit, and because the prompt prefix through step eight is unchanged, the KV cache is reused over 95% on replay, so early steps aren't reprocessed again. Once the run can be forked, a meta-agent can sit on top and operate it. It watches the trace and reverts as soon as it looks wrong, before the bad write is committed. In practice, it's just Python calling fork, replay, and revert on the trace, rather than a separate control plane wired into the harness. Not everything is reversible though. Files and sandbox changes undo themselves, but a database write has no automatic undo, so it needs a matching undo step set up in advance. Something external, like a sent email or a real charge, can't be undone, so the supervisor's job there is to catch it before it fires. They tested this on a few public benchmarks. On CooperBench, where two agents work on the same codebase, adding a live supervisor took the pair-coding pass rate from 28.8% to 54.7%. It's still early and labeled alpha. The benefit mostly shows up when a run gets branched a lot over a heavy sandbox state, which is exactly where restarting wastes the most tokens and time. If Git was made to make file changes reversible, Shepherd is trying to do the same thing for a live agent run. Shepherd Repo: (don't forget to star it ⭐ ) That said, Shepherd reverts a bad step inside a run. The harness around it, the prompts, tools, and checks the supervisor relies on, still drifts across runs as models and dependencies change. Akshay wrote about making that harness repair itself, where a failing trace gets diagnosed, the fix is verified against the exact input that failed, and the failure is locked as a regression test so it can't recur. Read it below.

Avi Chawla

440,706 次观看 • 1 个月前

One prompt. 6 frontier coding models. "Create a realistic fireworks show using HTML Canvas and JavaScript. No libraries." Some built a whole celebration. Others... lit a sparkler. The lineup: - GPT-5.3 Codex - Claude Opus 4.6 - Gemini 3.1 Pro - MiniMax M2.7 - GLM-5 - Kimi K2.5
0:14

One prompt. 6 frontier coding models. "Create a realistic fireworks show using HTML Canvas and JavaScript. No libraries." Some built a whole celebration. Others... lit a sparkler. The lineup: - GPT-5.3 Codex - Claude Opus 4.6 - Gemini 3.1 Pro - MiniMax M2.7 - GLM-5 - Kimi K2.5

stevibe

14,152 次观看 • 5 个月前

VTubing is for everyone! I don't like to bring this up, but recent events in the VTuber community made some people really vile. I had my M&G at HolMat last week and had people come up and just yell at me for being a girl in the "womans world", that is VTubing, no idea about who I was, or why I was there. I had the same group return to my handler multiple times, I felt really worried about the person carrying me around having to deal with this repeatedly and tried to steer them away from the group. It seems they found my YT and left me a handful of the same comments, luckily all caught by moderation tools. While I will refrain from M&G's for a little until this calms down I want to say this; I have been managing for over 1.5 years now, I deliberately take on male VTubers to show them that they can still do it. I take on people that have babiniku accounts, those that are changing from one gender presentation to another and want help, and those that have no gender presentation in their avatar. A lot of male VTubers struggle to find a manager because the stereotype is that all male VTubers are evil and it perpetuates a stereotype that extends beyond entertainment industry subcultures like streaming. VTubing has always been a medium, while character and marketing matters, it's about what makes you happy, YOU are the person that makes the content. Use whatever you want as an avatar, use a voice changer if that's what you want to do. Every month I have a male client bring up they can't do it the same way, that it's easier if jiggle physics or a cute voice is the answer to fast growth. And I tell them yeah, I can also do ASMR, I can do drama content, collab with a larger person, and get to x amount of viewers. There are shortcuts in every entertainment profession and subculture. Does it last? Do quick fixes for anything ever last? Yes, great physics and an expensive model can get people in, but if your value, your way of interacting with people, your content plan and marketing is ass, you can pack up. Most of us start on a small budget, premade or resold models, and it's the same in the big league entertainment industry too. Can I take out a loan and put myself on a billboard tomorrow, can I pay the most expensive model artist and rigger and get in along corporate VTubers tomorrow? Sure. Will it last, will it be genuine, will people trust me? Hell no. Genuine communities and growth build trust. I think streaming, creative industries, entertainment, are full of people, regardless of gender, that will see success and call it "easy", because of what they see as the end product. They don't see most VTubers working a second job, they don't see the managers, they don't see 100+ hours a month going into content production, years invested in singing and voice lessons, model redebut after redebut. I think a lot of male VTubers get a bad rep, because so many boys are raised without putting emphasis on empathy and creativity, watching my brothers be told they should not pursue art, that theatre class is a waste of time, and that they needed to go study x or y to make money for their families in the future was heartbreaking. Nobody should have to look at others and feel so much hatred for society they turn against a whole group of people. If you want to pursue entertainment, please do. If you want to grow and try and make a name for yourself, you should start there, not with yelling about how you already failed. "Oh but I can't", "Oh but the odds are stacked against me", look at the big streamers, look at Ironmouse who overcame everything with hard work, look at Kiara who rose from the ashes, look at everyone that fights against their odds every day and give it your all. If you already think you lost, then you have nothing to lose. Don't give up on your dreams because others tell you to.
0:15

VTubing is for everyone! I don't like to bring this up, but recent events in the VTuber community made some people really vile. I had my M&G at HolMat last week and had people come up and just yell at me for being a girl in the "womans world", that is VTubing, no idea about who I was, or why I was there. I had the same group return to my handler multiple times, I felt really worried about the person carrying me around having to deal with this repeatedly and tried to steer them away from the group. It seems they found my YT and left me a handful of the same comments, luckily all caught by moderation tools. While I will refrain from M&G's for a little until this calms down I want to say this; I have been managing for over 1.5 years now, I deliberately take on male VTubers to show them that they can still do it. I take on people that have babiniku accounts, those that are changing from one gender presentation to another and want help, and those that have no gender presentation in their avatar. A lot of male VTubers struggle to find a manager because the stereotype is that all male VTubers are evil and it perpetuates a stereotype that extends beyond entertainment industry subcultures like streaming. VTubing has always been a medium, while character and marketing matters, it's about what makes you happy, YOU are the person that makes the content. Use whatever you want as an avatar, use a voice changer if that's what you want to do. Every month I have a male client bring up they can't do it the same way, that it's easier if jiggle physics or a cute voice is the answer to fast growth. And I tell them yeah, I can also do ASMR, I can do drama content, collab with a larger person, and get to x amount of viewers. There are shortcuts in every entertainment profession and subculture. Does it last? Do quick fixes for anything ever last? Yes, great physics and an expensive model can get people in, but if your value, your way of interacting with people, your content plan and marketing is ass, you can pack up. Most of us start on a small budget, premade or resold models, and it's the same in the big league entertainment industry too. Can I take out a loan and put myself on a billboard tomorrow, can I pay the most expensive model artist and rigger and get in along corporate VTubers tomorrow? Sure. Will it last, will it be genuine, will people trust me? Hell no. Genuine communities and growth build trust. I think streaming, creative industries, entertainment, are full of people, regardless of gender, that will see success and call it "easy", because of what they see as the end product. They don't see most VTubers working a second job, they don't see the managers, they don't see 100+ hours a month going into content production, years invested in singing and voice lessons, model redebut after redebut. I think a lot of male VTubers get a bad rep, because so many boys are raised without putting emphasis on empathy and creativity, watching my brothers be told they should not pursue art, that theatre class is a waste of time, and that they needed to go study x or y to make money for their families in the future was heartbreaking. Nobody should have to look at others and feel so much hatred for society they turn against a whole group of people. If you want to pursue entertainment, please do. If you want to grow and try and make a name for yourself, you should start there, not with yelling about how you already failed. "Oh but I can't", "Oh but the odds are stacked against me", look at the big streamers, look at Ironmouse who overcame everything with hard work, look at Kiara who rose from the ashes, look at everyone that fights against their odds every day and give it your all. If you already think you lost, then you have nothing to lose. Don't give up on your dreams because others tell you to.

Kuromiya Lucien

15,552 次观看 • 7 个月前

The “I never wanted any of this to be public or content” Myth I never wanted any of this to come out. Except I actually mean that. It’s embarrassing, it’s trashy, and it’s the opposite of how I want to present myself publicly. I stayed silent for months while lie after lie was spun about me behind the scenes because I didn’t want to air private, messy moments from a relationship I genuinely cared about. I stayed silent out of fear, because I was told no one would believe me because he would be blindly believed on his large platform. In one of our very last conversations before I blocked him everywhere mid-October, he threatened to do a show on me, contact my employer, and “turn me into the next Lindsey”. Why? Because I didn’t want to continue apologizing to him for the 300th time about venting to someone I thought was a friend. I stayed silent because of these threats. But after Wednesday, after Chelsea posted the private video Aidan gave her, (the same one he threatened me with directly back on 10/2), and after the gaslighting posts yesterday, I don’t have a choice anymore. So here is the timeline. Here are the receipts. Here is what actually happened. September-October 2025- I Stayed Silent Through Months of Smear Campaigns For months, several creators repeated the same false narrative that “Aidan never said a bad word about Meredith.” Jess. Glarer. Auntie Deb. Kim. Ray from Dallas. Others. Meanwhile, Aidan was: • Badmouthing me privately and encouraging others to do so publicly, • Texting Joe “Flipperhead” entire paragraphs trashing me, • Aidan knew Joe would leak them (and even bragged on a stream that he did exactly that so he’d leak them), • Sending Jess Machado after me for months on her large platform, • Directing people behind the scenes to paint me as a villain. • Sending random people who supported me DMs to change their mind (yes Aidan, not everyone is fooled by your bullshit and they came right to me). I ignored all of it. I said nothing. I stayed silent through daily lies because I didn’t want to call more attention to it or face Aidan’s wrath. November 2025 -The Rumors About “March 4”, Joe Flipperhead Flips to Team Aidan Aidan texted Joe his entire made-up version weeks ago. Joe circulated those texts on purpose. Joe threatened me with posting a video montage of mine and Aidan’s private texts. Joe is a pussy so he eventually had Kristy post it for him. Creators discussed it publicly without me ever responding. I STILL didn’t defend myself. November 26-27, 2025- Ratchet Chelsea: The Full 48 Hour Meltdown 11/26/25- The Unprovoked Attack & The March 4 Video On Tuesday night, Chelsea came out of nowhere and launched herself straight at me, accusing me of all kinds of conspiracies from harassment to the horrific crime of sending a friend request, which is explained in vivid detail in the below post (she’s since gone somewhat viral, love this for her!) She spent the next 24 hours thoroughly embarrassing herself and the moment anyone challenged her story even slightly, she did what she always does: she imploded. Her accusations spiraled into her posting the March 4 video. On 11/26/25 at 11:08 PM, Chelsea posted the first clip–a blanket, context-less snippet of me drunk saying something I immediately apologized for the next morning. At 11:50 PM, I responded because at that point, I had no choice. 11/27/25-The Gaslighting & 9-Minute Backpedal Then yesterday, Aidan tried to save face by posting that he “didn’t want the video to be public.” And like clockwork, nine minutes later, Chelsea came charging into the comments like a Temu-sponsored bat out of hell, scrambling to apologize, trying to rewrite reality so it looked like he didn’t send her to do it. It was panicked, sloppy damage control. An amateur quality cover-up attempt that only made it more obvious how coordinated this all was. She wasn’t apologizing because she grew a conscience. She was apologizing because she got caught doing exactly what she’d been primed to do. I wonder if Aidan called her “you should’ve waited at least 20 minutes, dumbass!!!”. Not too bright that Chelsea. I can only imagine the regrets he must have for utilizing someone with the IQ of a pencil to do his dirty work. The Coordination Between Aidan & Chelsea Is Obvious Chelsea had virtually NO followers Wednesday morning. Aidan was one of the first. We’re supposed to believe someone with 150k followers just stumbled on an account that starts posting HIS private messages and HIS private videos? Kk. Aidan commented on her posts that same day, clearly encouraging it. She confirmed in my DMs back on 11/6 that she and Aidan were aligned. And she literally wrote at 4:04 AM that she had “proof,” videos & screenshots she would release if I didn’t “stop calling” her, calls she still hasn’t produced a single shred of proof of. Then yesterday within 9 minutes of his post, she writes: “Sorry for going rogue, Aidan, I hope you don’t hate my guts.” Going rogue? From WHAT? Who gave you the material in the first place? The coordination is obvious to anyone with functioning brain cells. The “I never wanted this public” lie Aidan’s post yesterday claimed he “never wanted this to be public.” If this were true, why did he instruct Chelsea to post it? Why did he share it with Jess Machado to threaten me with for the last month? Why did he share it with Joe Flipperhead, and God knows who else? If he didn’t want it public, he wouldn’t have shared it with anyone, let alone the girl I caught him cheating with. The timeline is clear– Chelsea posted a video at 11:08 pm on Thanksgiving Eve, I responded at 11:50 out of self defense. He leaked first. He escalated first. He weaponized it first. Trying to rewrite that now is gaslighting, pure and simple. He's been lighting matches behind the scenes for months, praying for this to go public. The “Meredith accused him of abuse” narrative is FALSE My actions tell the truth: • I never went to police. • I never filed anything. • I never told anyone he hit me. • I never repeated it. • I apologized the next day. • We stayed together for FOUR more months after that night. • He called me 100+ times some days, even indicating he’d end his life if we broke up. If he believed I was “dangerous,” his actions would have said that. They didn’t. The only reason this is public now is because HE leaked it. Actions > drunk slurred words said because a camera was shoved in my face when I asked him repeatedly to stop filming and to leave. The Wilbur Theatre Rumors—Let’s get petty for a second. This is another topic I’m cringing at discussing but we can thank Aidan for since he shared our sexual messages with Joe Flipperhead. And since some people (Auntie Deb, sweetie, this means you) insist on pushing their Dollar Store fan-fiction about the Wilbur Theatre night, let’s actually walk through what happened, using facts instead of whatever drug-induced hallucinations you’ve been spinning this week. The rumor goes like this- I was “mad at Aidan because we didn’t hook up,” and to “prove” it, they trot out a cherry-picked text where I said I wanted to hook up, he didn’t answer, and the next morning I said I was drunk and sorry. And somehow this has been spun into me being desperate, obsessed, or pining like a background character in a teenage soap opera. Adorable. Wrong, but adorable. Here’s the real plot twist-Aidan and I had already been together literally two days earlier, and shocker-that was initiated by him, not me. I didn’t just “show up at the Wilbur”. He invited me to come to the after party at Encore knowing I had a wedding earlier in the night. I wanted to see my friends there who I actually met to go with. This wasn’t a surprise, it was planned ahead of time. The “I was drunk, sorry” message wasn’t heartbreak. It wasn’t longing. It was me politely tapping the brakes because I didn’t even want to entertain whatever bad path it could lead us back down. And then? He texted me asking about my brother’s wedding, how I am, etc. I ignored him for a full week. Until I had no choice but to speak to him about MereNeill. That silence, from me, is what triggered his latest meltdown. Not mine. His. So no, Auntie Deb, your version isn’t “a different perspective.” It’s just wrong. Like wildly, hilariously, not-even-in-the-ballpark wrong. But sure, keep spinning fanfic if it makes your livestreams feel more exciting. I realize it’s probably been awhile since you’ve had any action since you’ve let yourself go so badly, so you live vicariously through others. Recap For MONTHS I ignored: • the texts he sent Joe to leak • the smear campaign • the creators parroting lies • the behind-the-scenes messages • the insinuations I hacked him and even shared revenge porn • the constant coordinated attacks • the threats of “turning me into the next Lindsey” • his warning that everyone would believe him because of his platform • the threats to contact my employer I didn’t clap back, didn’t retaliate, didn’t respond. But Wednesday, when Chelsea dumped the video he gave her, the same video she threatened weeks ago to drop, AND he posted pretending he didn’t want this all out? That was the line. They chose the nuclear option. Now I’m responding with facts in pure self-defense. Conclusion/Message for Temu Storm This entire mess could have stayed private. I wanted it to stay private. But when someone leaks your private moments, lies about your intentions, weaponizes your past, and coordinates an online attack, you either let the false narrative stand or you defend yourself. I’m choosing to defend myself with receipts, not gossip. With timelines, not “he said.” With evidence, not weaponized drunk clips. And then there’s Aidan’s favorite delusion. The claim that I was ‘working with Karen behind the scenes’ or ‘conspiring with her to get him in trouble for the recording.’ That could not be further from the truth. I didn’t even know he had recorded her until after she already knew about it herself. I wasn’t working with her, plotting with her, or communicating with her about it. She and I have always been friendly, but we didn’t even discuss the recording until it started leaking and Joe Flipperhead went feral on Twitter. Meanwhile, Aidan spent weeks screaming at me, accusing me of being some kind of secret double agent, like he was trying to create the betrayal he was terrified of. Almost a self-fulfilling prophecy. And here’s the reality: in the last few weeks, after he’s gone fully scorched earth on me, I have talked to her, and I support her 100%. The things he’s done to her over the past couple years, and the way he twisted it all for his followers, is the exact same manipulation I lived through. I’m not going into her details because that’s her story to tell… but let’s just say I have a feeling you’ll be hearing it sooner rather than later. And since this week has apparently become “Let’s All Fixate on Meredith Week,” let me address the content-creator sideshow, too. Some of you are strangers, while some I actually considered my friends. It’s shocking but not shocking how quickly you all flip a switch and follow your captain’s orders. Ray from Dallas spent his Sunday foaming at the mouth in a hostile little video about me, all bark, no substance. I saw it. I’m not intimidated. Mostly, I’m embarrassed for you. And your family(ies) will feel the same when your ass is doxxed next week. Can’t wait to tAkE tHaT WaLk with you! Then there’s Will, who’s gone on multiple streams calling me “crazy” and pretending he’s scared of me, despite me being nothing but nice to him. Why? Because he thinks I was the first one to tell Karen he heard the recording, which he absolutely did (unless Aidan is lying about who he played it for, but we know he’s actually telling the truth on this for once). She was informed of it the very next day, which is why she called you and you know this. This was two full days before I was even aware of a call at all. Will, for someone with your track record, maybe sit the “women are dangerous” narrative out. You’re not fooling anyone. I’m staring at 15 pages of police reports and no, it’s not all “things you’ve owned up for in the past”. You know this. Again, I’d take a very large seat here. And then my personal favorite-ex-fake friend Auntie Deb, who decided to spice up his Spaces on Thanksgiving by accusing me of distributing revenge porn with zero evidence, zero screenshots, zero anything. A completely fabricated felony tossed around like it’s gossip hour at the bingo hall. Considering your own professional history, James, I’d hope you of all people would understand how catastrophic false accusations can be, but apparently not. It would be a shame if I were to share the real reason around your separation at the middle school and how you wanted to bring Turtlenoy into it. This is all based on the several witnesses I’ve spoken to. See how that works? You announce things as fact based on something someone told you. I wonder if you’ll have that same enthusiasm when the topic is you. TBD. I’ll wait for that retraction about revenge porn. Ball’s in your court hun. And of course, there’s Jessica Machado, who’s been grinding this axe for months like it’s her full-time personality. Hos long did you cry when Chelsea beat you to the punch with that video? Don’t worry hun, there’s more but she didn’t post it because it’s not great for your fairytale. If you thought the fall-out of Kate’s video was bad, Jessica…stay tuned. I may have been momentarily distracted dealing with the chaos from Temu Chelsea, but there’s an army of your victims out there that are eager and ready for your downfall. The false claim that I shared “revenge porn,” the conspiracy theories, the wild accusations you present as fact, all of it. The 10+ streams you’ve now defamed me on. You’ve been so obsessed with me for so long that you can’t even keep your own narratives straight anymore. I love this so much for you. Let me be crystal clear–I am done being all of your punching bag. You love to spin this narrative that I’m this dangerous person. You’d think you’d lay off from constantly f*cking with me if you truly believed that. I would genuinely love for this nonsense to stop. But that requires ALL of you to stop manufacturing drama, stop lying, and stop weaponizing made-up crimes for clicks. And if they want to keep going? I’ve got plenty of content for many seasons to come.
0:57

The “I never wanted any of this to be public or content” Myth I never wanted any of this to come out. Except I actually mean that. It’s embarrassing, it’s trashy, and it’s the opposite of how I want to present myself publicly. I stayed silent for months while lie after lie was spun about me behind the scenes because I didn’t want to air private, messy moments from a relationship I genuinely cared about. I stayed silent out of fear, because I was told no one would believe me because he would be blindly believed on his large platform. In one of our very last conversations before I blocked him everywhere mid-October, he threatened to do a show on me, contact my employer, and “turn me into the next Lindsey”. Why? Because I didn’t want to continue apologizing to him for the 300th time about venting to someone I thought was a friend. I stayed silent because of these threats. But after Wednesday, after Chelsea posted the private video Aidan gave her, (the same one he threatened me with directly back on 10/2), and after the gaslighting posts yesterday, I don’t have a choice anymore. So here is the timeline. Here are the receipts. Here is what actually happened. September-October 2025- I Stayed Silent Through Months of Smear Campaigns For months, several creators repeated the same false narrative that “Aidan never said a bad word about Meredith.” Jess. Glarer. Auntie Deb. Kim. Ray from Dallas. Others. Meanwhile, Aidan was: • Badmouthing me privately and encouraging others to do so publicly, • Texting Joe “Flipperhead” entire paragraphs trashing me, • Aidan knew Joe would leak them (and even bragged on a stream that he did exactly that so he’d leak them), • Sending Jess Machado after me for months on her large platform, • Directing people behind the scenes to paint me as a villain. • Sending random people who supported me DMs to change their mind (yes Aidan, not everyone is fooled by your bullshit and they came right to me). I ignored all of it. I said nothing. I stayed silent through daily lies because I didn’t want to call more attention to it or face Aidan’s wrath. November 2025 -The Rumors About “March 4”, Joe Flipperhead Flips to Team Aidan Aidan texted Joe his entire made-up version weeks ago. Joe circulated those texts on purpose. Joe threatened me with posting a video montage of mine and Aidan’s private texts. Joe is a pussy so he eventually had Kristy post it for him. Creators discussed it publicly without me ever responding. I STILL didn’t defend myself. November 26-27, 2025- Ratchet Chelsea: The Full 48 Hour Meltdown 11/26/25- The Unprovoked Attack & The March 4 Video On Tuesday night, Chelsea came out of nowhere and launched herself straight at me, accusing me of all kinds of conspiracies from harassment to the horrific crime of sending a friend request, which is explained in vivid detail in the below post (she’s since gone somewhat viral, love this for her!) She spent the next 24 hours thoroughly embarrassing herself and the moment anyone challenged her story even slightly, she did what she always does: she imploded. Her accusations spiraled into her posting the March 4 video. On 11/26/25 at 11:08 PM, Chelsea posted the first clip–a blanket, context-less snippet of me drunk saying something I immediately apologized for the next morning. At 11:50 PM, I responded because at that point, I had no choice. 11/27/25-The Gaslighting & 9-Minute Backpedal Then yesterday, Aidan tried to save face by posting that he “didn’t want the video to be public.” And like clockwork, nine minutes later, Chelsea came charging into the comments like a Temu-sponsored bat out of hell, scrambling to apologize, trying to rewrite reality so it looked like he didn’t send her to do it. It was panicked, sloppy damage control. An amateur quality cover-up attempt that only made it more obvious how coordinated this all was. She wasn’t apologizing because she grew a conscience. She was apologizing because she got caught doing exactly what she’d been primed to do. I wonder if Aidan called her “you should’ve waited at least 20 minutes, dumbass!!!”. Not too bright that Chelsea. I can only imagine the regrets he must have for utilizing someone with the IQ of a pencil to do his dirty work. The Coordination Between Aidan & Chelsea Is Obvious Chelsea had virtually NO followers Wednesday morning. Aidan was one of the first. We’re supposed to believe someone with 150k followers just stumbled on an account that starts posting HIS private messages and HIS private videos? Kk. Aidan commented on her posts that same day, clearly encouraging it. She confirmed in my DMs back on 11/6 that she and Aidan were aligned. And she literally wrote at 4:04 AM that she had “proof,” videos & screenshots she would release if I didn’t “stop calling” her, calls she still hasn’t produced a single shred of proof of. Then yesterday within 9 minutes of his post, she writes: “Sorry for going rogue, Aidan, I hope you don’t hate my guts.” Going rogue? From WHAT? Who gave you the material in the first place? The coordination is obvious to anyone with functioning brain cells. The “I never wanted this public” lie Aidan’s post yesterday claimed he “never wanted this to be public.” If this were true, why did he instruct Chelsea to post it? Why did he share it with Jess Machado to threaten me with for the last month? Why did he share it with Joe Flipperhead, and God knows who else? If he didn’t want it public, he wouldn’t have shared it with anyone, let alone the girl I caught him cheating with. The timeline is clear– Chelsea posted a video at 11:08 pm on Thanksgiving Eve, I responded at 11:50 out of self defense. He leaked first. He escalated first. He weaponized it first. Trying to rewrite that now is gaslighting, pure and simple. He's been lighting matches behind the scenes for months, praying for this to go public. The “Meredith accused him of abuse” narrative is FALSE My actions tell the truth: • I never went to police. • I never filed anything. • I never told anyone he hit me. • I never repeated it. • I apologized the next day. • We stayed together for FOUR more months after that night. • He called me 100+ times some days, even indicating he’d end his life if we broke up. If he believed I was “dangerous,” his actions would have said that. They didn’t. The only reason this is public now is because HE leaked it. Actions > drunk slurred words said because a camera was shoved in my face when I asked him repeatedly to stop filming and to leave. The Wilbur Theatre Rumors—Let’s get petty for a second. This is another topic I’m cringing at discussing but we can thank Aidan for since he shared our sexual messages with Joe Flipperhead. And since some people (Auntie Deb, sweetie, this means you) insist on pushing their Dollar Store fan-fiction about the Wilbur Theatre night, let’s actually walk through what happened, using facts instead of whatever drug-induced hallucinations you’ve been spinning this week. The rumor goes like this- I was “mad at Aidan because we didn’t hook up,” and to “prove” it, they trot out a cherry-picked text where I said I wanted to hook up, he didn’t answer, and the next morning I said I was drunk and sorry. And somehow this has been spun into me being desperate, obsessed, or pining like a background character in a teenage soap opera. Adorable. Wrong, but adorable. Here’s the real plot twist-Aidan and I had already been together literally two days earlier, and shocker-that was initiated by him, not me. I didn’t just “show up at the Wilbur”. He invited me to come to the after party at Encore knowing I had a wedding earlier in the night. I wanted to see my friends there who I actually met to go with. This wasn’t a surprise, it was planned ahead of time. The “I was drunk, sorry” message wasn’t heartbreak. It wasn’t longing. It was me politely tapping the brakes because I didn’t even want to entertain whatever bad path it could lead us back down. And then? He texted me asking about my brother’s wedding, how I am, etc. I ignored him for a full week. Until I had no choice but to speak to him about MereNeill. That silence, from me, is what triggered his latest meltdown. Not mine. His. So no, Auntie Deb, your version isn’t “a different perspective.” It’s just wrong. Like wildly, hilariously, not-even-in-the-ballpark wrong. But sure, keep spinning fanfic if it makes your livestreams feel more exciting. I realize it’s probably been awhile since you’ve had any action since you’ve let yourself go so badly, so you live vicariously through others. Recap For MONTHS I ignored: • the texts he sent Joe to leak • the smear campaign • the creators parroting lies • the behind-the-scenes messages • the insinuations I hacked him and even shared revenge porn • the constant coordinated attacks • the threats of “turning me into the next Lindsey” • his warning that everyone would believe him because of his platform • the threats to contact my employer I didn’t clap back, didn’t retaliate, didn’t respond. But Wednesday, when Chelsea dumped the video he gave her, the same video she threatened weeks ago to drop, AND he posted pretending he didn’t want this all out? That was the line. They chose the nuclear option. Now I’m responding with facts in pure self-defense. Conclusion/Message for Temu Storm This entire mess could have stayed private. I wanted it to stay private. But when someone leaks your private moments, lies about your intentions, weaponizes your past, and coordinates an online attack, you either let the false narrative stand or you defend yourself. I’m choosing to defend myself with receipts, not gossip. With timelines, not “he said.” With evidence, not weaponized drunk clips. And then there’s Aidan’s favorite delusion. The claim that I was ‘working with Karen behind the scenes’ or ‘conspiring with her to get him in trouble for the recording.’ That could not be further from the truth. I didn’t even know he had recorded her until after she already knew about it herself. I wasn’t working with her, plotting with her, or communicating with her about it. She and I have always been friendly, but we didn’t even discuss the recording until it started leaking and Joe Flipperhead went feral on Twitter. Meanwhile, Aidan spent weeks screaming at me, accusing me of being some kind of secret double agent, like he was trying to create the betrayal he was terrified of. Almost a self-fulfilling prophecy. And here’s the reality: in the last few weeks, after he’s gone fully scorched earth on me, I have talked to her, and I support her 100%. The things he’s done to her over the past couple years, and the way he twisted it all for his followers, is the exact same manipulation I lived through. I’m not going into her details because that’s her story to tell… but let’s just say I have a feeling you’ll be hearing it sooner rather than later. And since this week has apparently become “Let’s All Fixate on Meredith Week,” let me address the content-creator sideshow, too. Some of you are strangers, while some I actually considered my friends. It’s shocking but not shocking how quickly you all flip a switch and follow your captain’s orders. Ray from Dallas spent his Sunday foaming at the mouth in a hostile little video about me, all bark, no substance. I saw it. I’m not intimidated. Mostly, I’m embarrassed for you. And your family(ies) will feel the same when your ass is doxxed next week. Can’t wait to tAkE tHaT WaLk with you! Then there’s Will, who’s gone on multiple streams calling me “crazy” and pretending he’s scared of me, despite me being nothing but nice to him. Why? Because he thinks I was the first one to tell Karen he heard the recording, which he absolutely did (unless Aidan is lying about who he played it for, but we know he’s actually telling the truth on this for once). She was informed of it the very next day, which is why she called you and you know this. This was two full days before I was even aware of a call at all. Will, for someone with your track record, maybe sit the “women are dangerous” narrative out. You’re not fooling anyone. I’m staring at 15 pages of police reports and no, it’s not all “things you’ve owned up for in the past”. You know this. Again, I’d take a very large seat here. And then my personal favorite-ex-fake friend Auntie Deb, who decided to spice up his Spaces on Thanksgiving by accusing me of distributing revenge porn with zero evidence, zero screenshots, zero anything. A completely fabricated felony tossed around like it’s gossip hour at the bingo hall. Considering your own professional history, James, I’d hope you of all people would understand how catastrophic false accusations can be, but apparently not. It would be a shame if I were to share the real reason around your separation at the middle school and how you wanted to bring Turtlenoy into it. This is all based on the several witnesses I’ve spoken to. See how that works? You announce things as fact based on something someone told you. I wonder if you’ll have that same enthusiasm when the topic is you. TBD. I’ll wait for that retraction about revenge porn. Ball’s in your court hun. And of course, there’s Jessica Machado, who’s been grinding this axe for months like it’s her full-time personality. Hos long did you cry when Chelsea beat you to the punch with that video? Don’t worry hun, there’s more but she didn’t post it because it’s not great for your fairytale. If you thought the fall-out of Kate’s video was bad, Jessica…stay tuned. I may have been momentarily distracted dealing with the chaos from Temu Chelsea, but there’s an army of your victims out there that are eager and ready for your downfall. The false claim that I shared “revenge porn,” the conspiracy theories, the wild accusations you present as fact, all of it. The 10+ streams you’ve now defamed me on. You’ve been so obsessed with me for so long that you can’t even keep your own narratives straight anymore. I love this so much for you. Let me be crystal clear–I am done being all of your punching bag. You love to spin this narrative that I’m this dangerous person. You’d think you’d lay off from constantly f*cking with me if you truly believed that. I would genuinely love for this nonsense to stop. But that requires ALL of you to stop manufacturing drama, stop lying, and stop weaponizing made-up crimes for clicks. And if they want to keep going? I’ve got plenty of content for many seasons to come.

Meredith O

16,962 次观看 • 8 个月前

You Can't Vibe-Code Trust Avishai Abrahami, Co-Founder & CEO of Wix , interviewed by Harry Stebbings (kevin andres) Summary: Wix trades at a $2.8B market cap on $2.1B of revenue while the market ascribes roughly zero value to a business throwing off $400M a year in free cash flow. Wix CEO Avishai Abrahami's argument is that the market can't yet price what AI actually threatens: the moat is trust and business logic, and neither gets vibe-coded away. His response is to own the disruptor (Base44), train his own narrow models, and stay committed through a storm he insists always arrives on a random Wednesday. 1. Trust is the moat. The real value of Salesforce is trust: JP Morgan and huge banks let it hold all their customer data, and the CRM itself is a small part of that. "What other platform will JP Morgan trust for their customers' data? None." That trust took years to build and can't be reconstructed by an agent scraping a database, so the companies whose value lives in trust survive the SaaS apocalypse while the ones reduced to piping get commoditized. 2. The business-logic wall. "You're not going to vibe-code Shopify no matter how good you are. The business logic is too hard." Wix tested this directly: they asked a team of professional developers to build the operating logic for a single hairdresser in Base44, gave up after a week, brought in a stronger team, and still failed two weeks later. Complex operational software is far harder than a demo suggests, which is why the pizza shop and the hairdresser stay Wix customers rather than build their own stack. 3. Own the disruptor. Wix bought Base44, a one-person company, for $80M, and it now does over $150M in ARR, roughly double what they paid. Abrahami frames the future as three buckets: owners who never want to build, owners who vibe-code everything themselves, and a mix in the middle over the next five or six years. Rather than bet on which wins, Wix owns the tool customers would defect to, so a customer who switches platforms still switches to Wix. 4. Trading on someone else's news. "Today we are trading on other companies' news. We're not trading on Wix news. We're trading on what OpenAI or Anthropic or Google are saying." Base44 alone, valued on vibe-coding peer multiples, should be worth around $8B, which means the market assigns less than zero to Wix's core. Abrahami's response is to detach: he doesn't wake up checking whether the stock moved 20%, because the only thing he can influence is the business. 5. The narrow model. Wix fine-tuned and combined its own models and now matches top-tier frontier quality on Base44 tasks at far lower cost. The logic: they sit on a huge stream of training data from watching what users try and where they fail, so a model built for Base44 can skip what frontier models carry, like knowledge of Chinese poetry, and go deep on what someone means when they say "build me a task manager to tell my boyfriend where he's wrong." A narrow target is easier to hit than a frontier model, and Wix already runs a trained model on website generation that's faster, cheaper, and makes fewer errors, retrained weekly on a live feedback loop. 6. Quality before cost. When Harry cites Chamath's claim that open source runs 14-16x cheaper, Abrahami pushes back: that holds for small tasks, but for something as complex as Base44 the savings land at 5-10%, and his own model runs 1-30% cheaper than frontier, not the order of magnitude people assume. More to the point, this is the wrong time to chase cost: "20% more quality, 20% less cost, I'll go for the quality." It's a brand-new market that's just starting, and the job now is to make the product better. 7. The but is very big. "We all give too much credit for AI. It's amazing, it's incredible, it's super powerful, but the but is pretty big." He asked Claude to write a safety protocol and got six mandatory gates, then pushed back on each one and watched the model cave until only one survived, downgrading the rest from "must test" to "might want to look at later." We over-trust these systems, and that reflex, treating a Reddit post as equivalent to research published in Nature, is where the danger lives. 8. Customer support still breaks. Wix has 3,500 people and its single biggest department is customer support, serving 192 countries. They tried hard not to build their own AI support agent, tested many off-the-shelf products, and concluded flatly: "It doesn't work. We tried, we tried again, it didn't work." The gap between hyped AI support startups and what actually ships in production is the tell that the technology is earlier than the marketing, maybe five years from being different. 9. Buybacks as dividends. Wix had $1.5B sitting in the bank it couldn't put into a major acquisition because it was focused on the new product and Base44, so it bought back stock at a low price, with admittedly terrible short-term timing. Abrahami is unbothered: "The big question is where it's going to be in three years, not what happened in the last three months." He argues buybacks are a fantastic, underused tool, essentially a dividend to every shareholder, and companies should lean on them to balance stock-based compensation instead of endlessly diluting. 10. Execution, not finance. A low stock price makes M&A currency less valuable, but Abrahami says that's not his real constraint. Base44 was a one-person company; Wix had to build an entire company around it, staffing it with people pulled from the core. "I don't know how to do another one of those at the same time and have the same quality." The bottleneck on the next acquisition is execution capacity, not the balance sheet. 11. Chosen to be here. The one thing money buys beyond food security is freedom, and the deepest form of that freedom is knowing you're here by choice. "I'm here because I've chosen to be here. Nobody made me." He could move to Costa Rica or dance carnival in Brazil, and choosing to stay and run a public company through a crashing stock is where he finds his power. Money also made him more impatient and a bit lazier, and more rational because he's no longer deciding from fear. 12. The random Wednesday. Resilience starts with accepting the storm will come, because we assume that if yesterday was easy tomorrow will be too, and reality doesn't move in gentle slopes. "The worst thing that happens is probably some random thing on some random Wednesday. It's not something you get a lot of warning for." His anchor, borrowed from Babylon 5, is that you get there when you get there and the weapons you have are the weapons you have, so the only real question is whether you're doing the best you can with what you control.
0:15

You Can't Vibe-Code Trust Avishai Abrahami, Co-Founder & CEO of Wix , interviewed by Harry Stebbings (kevin andres) Summary: Wix trades at a $2.8B market cap on $2.1B of revenue while the market ascribes roughly zero value to a business throwing off $400M a year in free cash flow. Wix CEO Avishai Abrahami's argument is that the market can't yet price what AI actually threatens: the moat is trust and business logic, and neither gets vibe-coded away. His response is to own the disruptor (Base44), train his own narrow models, and stay committed through a storm he insists always arrives on a random Wednesday. 1. Trust is the moat. The real value of Salesforce is trust: JP Morgan and huge banks let it hold all their customer data, and the CRM itself is a small part of that. "What other platform will JP Morgan trust for their customers' data? None." That trust took years to build and can't be reconstructed by an agent scraping a database, so the companies whose value lives in trust survive the SaaS apocalypse while the ones reduced to piping get commoditized. 2. The business-logic wall. "You're not going to vibe-code Shopify no matter how good you are. The business logic is too hard." Wix tested this directly: they asked a team of professional developers to build the operating logic for a single hairdresser in Base44, gave up after a week, brought in a stronger team, and still failed two weeks later. Complex operational software is far harder than a demo suggests, which is why the pizza shop and the hairdresser stay Wix customers rather than build their own stack. 3. Own the disruptor. Wix bought Base44, a one-person company, for $80M, and it now does over $150M in ARR, roughly double what they paid. Abrahami frames the future as three buckets: owners who never want to build, owners who vibe-code everything themselves, and a mix in the middle over the next five or six years. Rather than bet on which wins, Wix owns the tool customers would defect to, so a customer who switches platforms still switches to Wix. 4. Trading on someone else's news. "Today we are trading on other companies' news. We're not trading on Wix news. We're trading on what OpenAI or Anthropic or Google are saying." Base44 alone, valued on vibe-coding peer multiples, should be worth around $8B, which means the market assigns less than zero to Wix's core. Abrahami's response is to detach: he doesn't wake up checking whether the stock moved 20%, because the only thing he can influence is the business. 5. The narrow model. Wix fine-tuned and combined its own models and now matches top-tier frontier quality on Base44 tasks at far lower cost. The logic: they sit on a huge stream of training data from watching what users try and where they fail, so a model built for Base44 can skip what frontier models carry, like knowledge of Chinese poetry, and go deep on what someone means when they say "build me a task manager to tell my boyfriend where he's wrong." A narrow target is easier to hit than a frontier model, and Wix already runs a trained model on website generation that's faster, cheaper, and makes fewer errors, retrained weekly on a live feedback loop. 6. Quality before cost. When Harry cites Chamath's claim that open source runs 14-16x cheaper, Abrahami pushes back: that holds for small tasks, but for something as complex as Base44 the savings land at 5-10%, and his own model runs 1-30% cheaper than frontier, not the order of magnitude people assume. More to the point, this is the wrong time to chase cost: "20% more quality, 20% less cost, I'll go for the quality." It's a brand-new market that's just starting, and the job now is to make the product better. 7. The but is very big. "We all give too much credit for AI. It's amazing, it's incredible, it's super powerful, but the but is pretty big." He asked Claude to write a safety protocol and got six mandatory gates, then pushed back on each one and watched the model cave until only one survived, downgrading the rest from "must test" to "might want to look at later." We over-trust these systems, and that reflex, treating a Reddit post as equivalent to research published in Nature, is where the danger lives. 8. Customer support still breaks. Wix has 3,500 people and its single biggest department is customer support, serving 192 countries. They tried hard not to build their own AI support agent, tested many off-the-shelf products, and concluded flatly: "It doesn't work. We tried, we tried again, it didn't work." The gap between hyped AI support startups and what actually ships in production is the tell that the technology is earlier than the marketing, maybe five years from being different. 9. Buybacks as dividends. Wix had $1.5B sitting in the bank it couldn't put into a major acquisition because it was focused on the new product and Base44, so it bought back stock at a low price, with admittedly terrible short-term timing. Abrahami is unbothered: "The big question is where it's going to be in three years, not what happened in the last three months." He argues buybacks are a fantastic, underused tool, essentially a dividend to every shareholder, and companies should lean on them to balance stock-based compensation instead of endlessly diluting. 10. Execution, not finance. A low stock price makes M&A currency less valuable, but Abrahami says that's not his real constraint. Base44 was a one-person company; Wix had to build an entire company around it, staffing it with people pulled from the core. "I don't know how to do another one of those at the same time and have the same quality." The bottleneck on the next acquisition is execution capacity, not the balance sheet. 11. Chosen to be here. The one thing money buys beyond food security is freedom, and the deepest form of that freedom is knowing you're here by choice. "I'm here because I've chosen to be here. Nobody made me." He could move to Costa Rica or dance carnival in Brazil, and choosing to stay and run a public company through a crashing stock is where he finds his power. Money also made him more impatient and a bit lazier, and more rational because he's no longer deciding from fear. 12. The random Wednesday. Resilience starts with accepting the storm will come, because we assume that if yesterday was easy tomorrow will be too, and reality doesn't move in gentle slopes. "The worst thing that happens is probably some random thing on some random Wednesday. It's not something you get a lot of warning for." His anchor, borrowed from Babylon 5, is that you get there when you get there and the weapons you have are the weapons you have, so the only real question is whether you're doing the best you can with what you control.

Gokul Rajaram

22,657 次观看 • 14 天前