Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

A curated, regularly updated list of academic papers on large language model systems, showing how research has shifted from serving single requests to optimizing whole agent trajectories. Explore it here:

GitHub Projects Community

339,296 subscribers

10,683 views • 14 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Will the world rely on a single #GenAI model? Don’t count on it. In this AI Academy episode, IBM Director of Research Darío Gil explores different approaches to #AI and how businesses can go from AI users to value creators. Watch it here:

Will the world rely on a single #GenAI model? Don’t count on it. In this AI Academy episode, IBM Director of Research Darío Gil explores different approaches to #AI and how businesses can go from AI users to value creators. Watch it here:

IBM

38,693 views • 2 years ago

Been exploring a new way to explore AI research papers to discover deeper insights. Agents are at the center of it. So far, I've built this little interactive artifact generator in my orchestrator to visualize things. This allows me to change views and insights (on-demand) from 100s of papers. Just scratching the surface here. More to share soon.

Been exploring a new way to explore AI research papers to discover deeper insights. Agents are at the center of it. So far, I've built this little interactive artifact generator in my orchestrator to visualize things. This allows me to change views and insights (on-demand) from 100s of papers. Just scratching the surface here. More to share soon.

elvis

143,127 views • 6 months ago

We took a 30B model and split it in two to write tokens in parallel instead of one at a time. Introducing Nemotron-Labs-TwoTower: a diffusion language model from NVIDIA Research adapted from Nemotron-3-Nano-30B-A3B. Here’s how it works: one half holds the context, the other writes the tokens, with both reusing the pretrained model instead of training a new one from scratch. We found it kept 98.7% of the original model’s quality at 2.42× faster generation.

We took a 30B model and split it in two to write tokens in parallel instead of one at a time. Introducing Nemotron-Labs-TwoTower: a diffusion language model from NVIDIA Research adapted from Nemotron-3-Nano-30B-A3B. Here’s how it works: one half holds the context, the other writes the tokens, with both reusing the pretrained model instead of training a new one from scratch. We found it kept 98.7% of the original model’s quality at 2.42× faster generation.

NVIDIA AI

762,762 views • 3 months ago

We are entering an extremely exciting era for open-weight models. Kimi K2.6 now feels like a top agentic model. I took it for a spin via Fireworks AI fast inference APIs. Kimi K2.6 has impressive agentic capabilities, design skills, and the ability to synthesize large amounts of information. I built a little Skill that produces survey papers on any AI research topic you want. (see example in the clip) You can use the skill to tell your agent to generate a survey on whatever topic and watch it go to work. The artifact was fully generated by Kimi.ai's Kimi K2.6. It's cheap and fast. Next step for me is to explore ways to continue integrating the capabilities of these models on use cases like automating my LLM knowledge bases and augmenting my agent memory capabilities. Stay tuned for more.

We are entering an extremely exciting era for open-weight models. Kimi K2.6 now feels like a top agentic model. I took it for a spin via Fireworks AI fast inference APIs. Kimi K2.6 has impressive agentic capabilities, design skills, and the ability to synthesize large amounts of information. I built a little Skill that produces survey papers on any AI research topic you want. (see example in the clip) You can use the skill to tell your agent to generate a survey on whatever topic and watch it go to work. The artifact was fully generated by Kimi.ai's Kimi K2.6. It's cheap and fast. Next step for me is to explore ways to continue integrating the capabilities of these models on use cases like automating my LLM knowledge bases and augmenting my agent memory capabilities. Stay tuned for more.

elvis

47,678 views • 5 months ago

$here's how to make money with jev on autopilot it's a super fast decision model, so it picks an answer in a fifth of a second and tells you how sure it is - put instant pricing on a business that has none - sell the same thing to bigger businesses for more - sell a file that decides what an agent may do - charge monthly to keep all of it running - costs a fraction of a cent per decision - runs on hermes or grok bot, on a schedule just wrote a whole guide on how you can set this up:$

here's how to make money with jev on autopilot it's a super fast decision model, so it picks an answer in a fifth of a second and tells you how sure it is - put instant pricing on a business that has none - sell the same thing to bigger businesses for more - sell a file that decides what an agent may do - charge monthly to keep all of it running - costs a fraction of a cent per decision - runs on hermes or grok bot, on a schedule just wrote a whole guide on how you can set this up:

Chris

90,674 views • 20 days ago

my grok bot monitors the internet 24/7 and builds systems that make money on autopilot - pulls the problems out of what people wrote - only surfaces what more than one source confirmed - scores it and puts it on the dashboard - spins up a new agent when i approve - that agent builds it and sells it with whop - shows me what needs approving everyday everything from finding the customer to selling them just wrote a whole guide on how you can set this up exactly:

my grok bot monitors the internet 24/7 and builds systems that make money on autopilot - pulls the problems out of what people wrote - only surfaces what more than one source confirmed - scores it and puts it on the dashboard - spins up a new agent when i approve - that agent builds it and sells it with whop - shows me what needs approving everyday everything from finding the customer to selling them just wrote a whole guide on how you can set this up exactly:

Chris

50,983 views • 1 month ago

Building a personal knowledge base for my agents is increasingly where I spend my time these days. Like Andrej Karpathy, I also use Obsidian for my MD vaults. What's different in my approach is that I curate research papers on a daily basis and have actually tuned a Skill for months to find high-signal, relevant papers. I was reviewing and curating papers manually for some time, but now it's all automated as it has gotten so good at capturing what I consider the best of the best. There are so many papers these days, so this is a big deal. You all get to benefit from that with the papers I feature in my timeline and on DAIR.AI. The papers are indexed using tobi lutke qmd cli tool (all of it in markdown files along with useful metadata). So good for semantic search and surfacing insights, unlike anything out there. I am a visual person, so I then started to experiment with how to leverage this personal knowledge base of research papers inside my new interactive artifact generator (mcp tools inside my agent orchestrator system). The result is what you see in the clip. 100s of papers with all sorts of insights visualized. I keep track of research papers daily, so believe me when I tell you that this system is absolutely insane at surfacing insights. This is the result of months of tinkering on how to index research and leverage agent automations for wikification and robust documentation. But this is just the beginning. The visual artifact (which is interactive too) can be changed dynamically as I please. I can prompt my agent to throw any data at it. I can add different views to the data. Different interactions. I feel like this is the most personalized research system I have ever built and used, and it's not even close. The knowledge that the agents are able to surface from this basic setup is already extremely useful as I experiment with new agentic engineering concepts. I feel like this knowledge layer and the higher-level ones I am working on will allow me to maximize other automation tools like autoresearch. The research is only as good as the research questions. And the research questions are only as good as the insights the agents have access to. Where I am spending time now is on how to make this more actionable. I am obsessed about the search problem here. The automations, autoresearch, ralph research loop (I built one months ago) are easier to build but are only as good as what you feed them. Work in progress. More updates soon. Back to building.

Building a personal knowledge base for my agents is increasingly where I spend my time these days. Like Andrej Karpathy, I also use Obsidian for my MD vaults. What's different in my approach is that I curate research papers on a daily basis and have actually tuned a Skill for months to find high-signal, relevant papers. I was reviewing and curating papers manually for some time, but now it's all automated as it has gotten so good at capturing what I consider the best of the best. There are so many papers these days, so this is a big deal. You all get to benefit from that with the papers I feature in my timeline and on DAIR.AI. The papers are indexed using tobi lutke qmd cli tool (all of it in markdown files along with useful metadata). So good for semantic search and surfacing insights, unlike anything out there. I am a visual person, so I then started to experiment with how to leverage this personal knowledge base of research papers inside my new interactive artifact generator (mcp tools inside my agent orchestrator system). The result is what you see in the clip. 100s of papers with all sorts of insights visualized. I keep track of research papers daily, so believe me when I tell you that this system is absolutely insane at surfacing insights. This is the result of months of tinkering on how to index research and leverage agent automations for wikification and robust documentation. But this is just the beginning. The visual artifact (which is interactive too) can be changed dynamically as I please. I can prompt my agent to throw any data at it. I can add different views to the data. Different interactions. I feel like this is the most personalized research system I have ever built and used, and it's not even close. The knowledge that the agents are able to surface from this basic setup is already extremely useful as I experiment with new agentic engineering concepts. I feel like this knowledge layer and the higher-level ones I am working on will allow me to maximize other automation tools like autoresearch. The research is only as good as the research questions. And the research questions are only as good as the insights the agents have access to. Where I am spending time now is on how to make this more actionable. I am obsessed about the search problem here. The automations, autoresearch, ralph research loop (I built one months ago) are easier to build but are only as good as what you feed them. Work in progress. More updates soon. Back to building.

elvis

467,812 views • 6 months ago

Someone told ClawdBot to build a 6-agent Polymarket trading system while they slept. 6 hours. Not a single question asked. Here's what it built on its own: > Monitoring agent running 24/7 — spots mispriced markets, writes to memory, pings Telegram instantly > Research agent parsing news, X, and macro data every morning before you check your phone > Trading agent reading the research memory and acting before the market catches up > All of it running on one Gateway, one config file, isolated per agent First week results: - 47 mispriced markets caught before Polymarket adjusted - 8-12c avg entry edge per position - Token cost dropped 3x, from $0.40 to $0.13 per request The whole system is plain .md text files. Change one line, the agent behaves differently. No deploy. No build. A BOT RESPONDS. AN AGENT EARNS. THIS IS WHAT AGENTIC TRADING ACTUALLY LOOKS LIKE.

Someone told ClawdBot to build a 6-agent Polymarket trading system while they slept. 6 hours. Not a single question asked. Here's what it built on its own: > Monitoring agent running 24/7 — spots mispriced markets, writes to memory, pings Telegram instantly > Research agent parsing news, X, and macro data every morning before you check your phone > Trading agent reading the research memory and acting before the market catches up > All of it running on one Gateway, one config file, isolated per agent First week results: - 47 mispriced markets caught before Polymarket adjusted - 8-12c avg entry edge per position - Token cost dropped 3x, from $0.40 to $0.13 per request The whole system is plain .md text files. Change one line, the agent behaves differently. No deploy. No build. A BOT RESPONDS. AN AGENT EARNS. THIS IS WHAT AGENTIC TRADING ACTUALLY LOOKS LIKE.

0xMarioNawfal

80,450 views • 7 months ago

💡 This Isn’t Your Average Automation! Think of it like your smartest intern, but one that: • Understands your goals • Operates 24/7 without fatigue • Strategizes how to reach them • Executes within defined parameters This is the evolution of enterprise automation. Not just rules and scripts, but adaptive, goal-driven systems that sense, reason, and act on their own. We call it an AI Agent. 🔗 Explore here:

💡 This Isn’t Your Average Automation! Think of it like your smartest intern, but one that: • Understands your goals • Operates 24/7 without fatigue • Strategizes how to reach them • Executes within defined parameters This is the evolution of enterprise automation. Not just rules and scripts, but adaptive, goal-driven systems that sense, reason, and act on their own. We call it an AI Agent. 🔗 Explore here:

AITECH CLOUD NETWORK

52,167 views • 1 year ago

Day 15 of learning AI bs from scratch > looked into seq2seq models > understood the encoder-decoder architecture > implemented a basic Seq2Seq model using RNNs > how the decoder generates outputs one step at a time > read couple more chapters of > understood the fixed-context bottleneck I updated the resource list blog, link below, it has more resources organized now. Fell asleep on the desk towards the morning, rweeting rn after waking up.

Day 15 of learning AI bs from scratch > looked into seq2seq models > understood the encoder-decoder architecture > implemented a basic Seq2Seq model using RNNs > how the decoder generates outputs one step at a time > read couple more chapters of > understood the fixed-context bottleneck I updated the resource list blog, link below, it has more resources organized now. Fell asleep on the desk towards the morning, rweeting rn after waking up.

Arham Amin

31,853 views • 2 months ago

GEMMA 4 26B ON AN RTX 4060 WITH A 248K TOKEN CONTEXT WINDOW 20 tokens per second and a context window so large you can feed it entire codebases, books and research papers in a single prompt this is not a cloud api and not a server rack, this is a regular consumer gpu running locally with llama.cpp and q4_k_xl quantization 248k context on an 8gb vram card was not supposed to be possible and here it is just running on someone’s desk the article below covers exactly which tools and configs make this kind of setup work in 2026 ↓

GEMMA 4 26B ON AN RTX 4060 WITH A 248K TOKEN CONTEXT WINDOW 20 tokens per second and a context window so large you can feed it entire codebases, books and research papers in a single prompt this is not a cloud api and not a server rack, this is a regular consumer gpu running locally with llama.cpp and q4_k_xl quantization 248k context on an 8gb vram card was not supposed to be possible and here it is just running on someone’s desk the article below covers exactly which tools and configs make this kind of setup work in 2026 ↓

leopardracer

56,865 views • 4 months ago

VERCEL + CLOUDFLARE + CLAUDE CODE all picked up Jev in 6 days... an AI that can't write a single word became the fastest adopted model on Vercel's AI Gateway Jev decides → big model thinks → code authorizes TypeSafe dropped it on September 15, founder Diogo Almeida worked on ChatGPT research at OpenAI it answers three ways: pick one option, rate it on a scale, or yes/no with a percentage "which agent next?" comes back as Researcher 71% / Coder 18% / Reviewer 11% Vercel says almost 13% of its paying teams tried it within the first 24 hours on Vercel's leaderboard it pulled about 26% of all requests but only 1.8% of tokens TypeSafe's own tests: up to 194x faster and 445x cheaper than regular models TypeSafe calls it a System One Model, named after Kahneman's Thinking, Fast and Slow your expensive model stops burning tokens on yes or no questions and only gets the hard ones 6 days, 110 community use cases, zero words written ↓

VERCEL + CLOUDFLARE + CLAUDE CODE all picked up Jev in 6 days... an AI that can't write a single word became the fastest adopted model on Vercel's AI Gateway Jev decides → big model thinks → code authorizes TypeSafe dropped it on September 15, founder Diogo Almeida worked on ChatGPT research at OpenAI it answers three ways: pick one option, rate it on a scale, or yes/no with a percentage "which agent next?" comes back as Researcher 71% / Coder 18% / Reviewer 11% Vercel says almost 13% of its paying teams tried it within the first 24 hours on Vercel's leaderboard it pulled about 26% of all requests but only 1.8% of tokens TypeSafe's own tests: up to 194x faster and 445x cheaper than regular models TypeSafe calls it a System One Model, named after Kahneman's Thinking, Fast and Slow your expensive model stops burning tokens on yes or no questions and only gets the hard ones 6 days, 110 community use cases, zero words written ↓

leopardracer

16,184 views • 8 days ago

Here it is. The US seems to have wanted a trade deal with Canada as a whole and not navigate local red tape such as language requirements. "They made changes, including threats on the French language and Quebec culture, and Canadian culture." How did it threaten the Quebec culture? What is the threat to Canadian culture in this trade deal considering we have whole parts of Canada that don't resemble anything of Canadian culture after the Liberals' mass immigration...is it the Dairy Supply Management?

Here it is. The US seems to have wanted a trade deal with Canada as a whole and not navigate local red tape such as language requirements. "They made changes, including threats on the French language and Quebec culture, and Canadian culture." How did it threaten the Quebec culture? What is the threat to Canadian culture in this trade deal considering we have whole parts of Canada that don't resemble anything of Canadian culture after the Liberals' mass immigration...is it the Dairy Supply Management?

Kirk Lubimov

90,438 views • 1 month ago

Managed to do something crazy with Halo. It now has realtime iMessage sync support *without needing a Mac*. Everything done locally on your iPhone using Shortcuts automation. The agent now has context about your conversations/SMS and it opens up a whole world of possibilities from knowing about people, travel, transactions, etc. Another world-first for Halo. I expect others to follow.

Managed to do something crazy with Halo. It now has realtime iMessage sync support without needing a Mac. Everything done locally on your iPhone using Shortcuts automation. The agent now has context about your conversations/SMS and it opens up a whole world of possibilities from knowing about people, travel, transactions, etc. Another world-first for Halo. I expect others to follow.

Tanmay

15,927 views • 1 month ago

I MADE MY AI AGENT 10X FASTER WITHOUT CHANGING THE MODEL not a smarter model, not a bigger context window, not another clever prompt the same kind of AI that designs vaccines for viruses we have not even met yet was spending two minutes opening the wrong files just to hand me a brief from three months ago the problem was never capability, it was the scaffolding that piled up around my agent by accident, folder by folder an agent does not think in your categories, it searches from scratch every single time, and your tidy human folders are a maze to it the fix was almost stupidly small, one index file at the root of each big folder and a few numbers in front of the folder names slowest task dropped from 2 minutes to 26 seconds, fastest ones hit 10, zero model changes capability is cheap when the scaffolding around it is broken the article breaks down the whole system in 15 minutes ↓

I MADE MY AI AGENT 10X FASTER WITHOUT CHANGING THE MODEL not a smarter model, not a bigger context window, not another clever prompt the same kind of AI that designs vaccines for viruses we have not even met yet was spending two minutes opening the wrong files just to hand me a brief from three months ago the problem was never capability, it was the scaffolding that piled up around my agent by accident, folder by folder an agent does not think in your categories, it searches from scratch every single time, and your tidy human folders are a maze to it the fix was almost stupidly small, one index file at the root of each big folder and a few numbers in front of the folder names slowest task dropped from 2 minutes to 26 seconds, fastest ones hit 10, zero model changes capability is cheap when the scaffolding around it is broken the article breaks down the whole system in 15 minutes ↓

shmidt

36,479 views • 3 months ago

🚨 Pakistan Propaganda Alert! A video showing a heavy MLRS (Multiple Launch Rocket Systems) barrage is being falsely shared as a real Pakistani attack on India. #PIBFactCheck ✅ The video is from a video game and has been online for over 3 years. ✅ It has no connection to the current India–Pakistan situation. 🔗Originally posted here: 📢 Stay alert. Verify before sharing! #IndiaFightsPropaganda

🚨 Pakistan Propaganda Alert! A video showing a heavy MLRS (Multiple Launch Rocket Systems) barrage is being falsely shared as a real Pakistani attack on India. #PIBFactCheck ✅ The video is from a video game and has been online for over 3 years. ✅ It has no connection to the current India–Pakistan situation. 🔗Originally posted here: 📢 Stay alert. Verify before sharing! #IndiaFightsPropaganda

PIB Fact Check

1,142,599 views • 1 year ago

I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on The summaries cost $3.99 on Together AI. The classifications cost $0.08 on TypeSafe AI. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live:

I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on The summaries cost $3.99 on Together AI. The classifications cost $0.08 on TypeSafe AI. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live:

Hassan

341,359 views • 23 days ago

HydraFusion Explained. Part I: How does the Copilot engine know what to optimize for? Your prompt is evaluated across 4 dimensions: ➡ Does it require deep reasoning? (aka. reasoning depth) ➡ Is it a sophisticated problem? (aka. code generation complexity) ➡ Is it untangling a complicated mess? (aka. debugging difficulty) ➡ Is it dominated by tool-use? (aka. tool orchestration needs) Based on this evaluation, a HyDRA score is assigned to determine the capability profile your task needs the most and to establish a quality bar. Part II: How does it choose a model? Note: It doesn' t pick one model to handle the entire job e2e, (that's Auto mode). Instead, it selects 1 of 3 execution workflows and assigns the best model at different stages based on the HyDRA score: 1️⃣ Single ⚙️ How it works: A single model completes the task from start to finish. ⚖️ Rationale: The task comfortably meets the quality bar with one model. Multi-model orchestration would add latency and cost with no meaningful quality gain. 2️⃣ Cascade ⚙️ How it works: A lightweight, cost-efficient model generates the solution. This draft is evaluated against a quality gate and if it falls short of the quality bar, the entire task escalates to a stronger, frontier model. ⚖️ Rationale: Only bring in the big guns when there is concrete evidence that a lightweight model won't meet the quality threshold. 3️⃣ Critique ⚙️ How it works: A lightweight model drafts the initial code and tool interactions. An independent, read-only frontier model reviews that draft and provides feedback. The original lightweight model then performs any targeted revision(s) before the final response is sent to the user. ⚖️ Rationale: Writing code (output tokens) is expensive while reviewing code (input tokens) is cheap. Instead of incurring the cost of a powerhouse writing hundreds of lines from scratch, a cost-efficient model writes the first draft, and the frontier model just reviews it and points out fixes. HydraFusion is available in experimental preview on the GitHub Copilot CLI: /experimental on, /model and select Hydrafusion (Research Preview)

Julia Muiruri

13,000 views • 1 month ago