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.

this is what a real AGI loop looks like: 6+ hours agent runtime 240M+ tokens 86%+ cache hit 500+ tool calls $125 spent = 1 output that would have taken an expert team months to complete and cost tens of thousands

Nathan Wilbanks

9,325 subscribers

58,330 views • 3 months 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

Who would have ever imagined that this Country would appoint a man that was a Heroin addict for 14 years to be the Secretary of Health and Human Services! This is what complete incompetence and insanity looks like.

Who would have ever imagined that this Country would appoint a man that was a Heroin addict for 14 years to be the Secretary of Health and Human Services! This is what complete incompetence and insanity looks like.

Suzie rizzio

92,704 views • 4 months ago

what if you can run an entire 0-person company — without the grind of running a team? matrix is the runtime that makes it possible. in last week’s limited beta, our users created tens of thousands of new 0-person companies and started real businesses in matrix. today, matrix is open to everyone. launch yours ↓

what if you can run an entire 0-person company — without the grind of running a team? matrix is the runtime that makes it possible. in last week’s limited beta, our users created tens of thousands of new 0-person companies and started real businesses in matrix. today, matrix is open to everyone. launch yours ↓

Matrix

528,326 views • 2 months ago

✨No auto-tune. No backing tracks. No fake fog, fireworks, or lasers. ❤️ Just an artist, a towel, a splash of water, and a piano. That was all it took to captivate tens of thousands of people. That’s what real talent looks like. 🎹🎶

✨No auto-tune. No backing tracks. No fake fog, fireworks, or lasers. ❤️ Just an artist, a towel, a splash of water, and a piano. That was all it took to captivate tens of thousands of people. That’s what real talent looks like. 🎹🎶

Love Music

22,240 views • 3 months ago

🇺🇸🇮🇷🇮🇱MARK LEVIN ON PRESIDENT TRUMP: "This is what a real leader looks like." There's a point in which you need to defend our country, defend our allies. And when you see a regime slaughtering tens of thousands of young people..

🇺🇸🇮🇷🇮🇱MARK LEVIN ON PRESIDENT TRUMP: "This is what a real leader looks like." There's a point in which you need to defend our country, defend our allies. And when you see a regime slaughtering tens of thousands of young people..

Donald J Trump Posts TruthSocial

35,577 views • 7 months ago

MARK LEVIN ON TRUMP: "This is what a real leader looks like." "There's a point in which you need to defend our country, defend our allies. And when you see a regime slaughtering tens of thousands of young people, shooting them in the street...to sit still there at a time like that is utterly and completely unconscionable." FOX & Friends

MARK LEVIN ON TRUMP: "This is what a real leader looks like." "There's a point in which you need to defend our country, defend our allies. And when you see a regime slaughtering tens of thousands of young people, shooting them in the street...to sit still there at a time like that is utterly and completely unconscionable." FOX & Friends

Fox News

283,806 views • 7 months ago

figured out how Cascadeur – The Future of Animation works. this is mocap i recorded on an #iPhone and converted to #UE5 skeleton with QuickMagic. been working on this pipeline for months, but this particular animation only took like 2 hours. quality like this would have cost $10,000 previously

figured out how Cascadeur – The Future of Animation works. this is mocap i recorded on an #iPhone and converted to #UE5 skeleton with QuickMagic. been working on this pipeline for months, but this particular animation only took like 2 hours. quality like this would have cost $10,000 previously

Andres Hernandez

14,243 views • 2 months ago

is insane. Can't tell if this is cool or really scary. It's an AI tool that sound like a REAL human and it's being used for sales calls. Air says it can deliver up to 40 minutes long with thousands of rep voices at users' disposal. Thoughts?

is insane. Can't tell if this is cool or really scary. It's an AI tool that sound like a REAL human and it's being used for sales calls. Air says it can deliver up to 40 minutes long with thousands of rep voices at users' disposal. Thoughts?

Taylor Peterson 💫

175,436 views • 2 years ago

"I hope [Diego Luna's] agent is a very, very busy human being the next couple hours, days, weeks and months because that is a young man this country would like to cheer for." More from our crew on tonight's Man of the Match for the U.S. Soccer Men's National Team.

"I hope [Diego Luna's] agent is a very, very busy human being the next couple hours, days, weeks and months because that is a young man this country would like to cheer for." More from our crew on tonight's Man of the Match for the U.S. Soccer Men's National Team.

FOX Soccer

57,139 views • 1 year ago

SpaceXAI engineer, Lauren Tan: "99% of people using GrokBot just for 1% of its real power. They run 1 agent without "loop" & "graph" I'm running a team of 20+ GrokBot agents, fully autonomous. I have a Chief of Staff agent, a PM agent and 20+ workers - that's the new stack of engineer" In a 1-hour session, a SpaceXAI engineer showed how to build a team of effective AI agents from scratch this is worth more than a $500 agentic engineering course watch this workshop today, then read how to build a fleet of GrokBot agents in the article below

SpaceXAI engineer, Lauren Tan: "99% of people using GrokBot just for 1% of its real power. They run 1 agent without "loop" & "graph" I'm running a team of 20+ GrokBot agents, fully autonomous. I have a Chief of Staff agent, a PM agent and 20+ workers - that's the new stack of engineer" In a 1-hour session, a SpaceXAI engineer showed how to build a team of effective AI agents from scratch this is worth more than a $500 agentic engineering course watch this workshop today, then read how to build a fleet of GrokBot agents in the article below

Codez

638,047 views • 24 days ago

SpaceXAI engineer, Lauren Tan: "99% of people using GrokBot just for 1% of its real power. They run 1 agent without "loop" & "graph" I'm running a team of 20+ GrokBot agents, fully autonomous. I have a Chief of Staff agent, a PM agent and 20+ workers - that's the new stack of engineer" In a 1-hour session, a SpaceXAI engineer showed how to build a team of effective AI agents from scratch this is worth more than a $500 agentic engineering course watch this workshop today, then read how to build a fleet of GrokBot agents in the article below

SpaceXAI engineer, Lauren Tan: "99% of people using GrokBot just for 1% of its real power. They run 1 agent without "loop" & "graph" I'm running a team of 20+ GrokBot agents, fully autonomous. I have a Chief of Staff agent, a PM agent and 20+ workers - that's the new stack of engineer" In a 1-hour session, a SpaceXAI engineer showed how to build a team of effective AI agents from scratch this is worth more than a $500 agentic engineering course watch this workshop today, then read how to build a fleet of GrokBot agents in the article below

0xRafy

47,726 views • 23 days ago

"I burned roughly $5,000 of API tokens in a few hours for $200. It's 100x subsidized. This is insanity" Kain on running 500+ agents to decompose an entire codebase on Claude Opus 4.8 "I set up my own personal account to test this on Opus 4.8. I know enough that I can come up with a task that's gonna burn a fuckload. In the course of four hours I used something like 200 million tokens and didn't hit any limits" "It was hundreds of agents, like 500 or 600 agents decomposing a codebase. It would have been somewhere on the order of $5,000 with API metered tokens, and I know how much tokens cost because that's where I live" "I ran this overnight multiple times. It's very possible I could have burned through $20,000 worth of API credits for $200. I knew it was subsidized, but it's subsidized to a level that's crazy. It's 100x subsidized"

"I burned roughly $5,000 of API tokens in a few hours for $200. It's 100x subsidized. This is insanity" Kain on running 500+ agents to decompose an entire codebase on Claude Opus 4.8 "I set up my own personal account to test this on Opus 4.8. I know enough that I can come up with a task that's gonna burn a fuckload. In the course of four hours I used something like 200 million tokens and didn't hit any limits" "It was hundreds of agents, like 500 or 600 agents decomposing a codebase. It would have been somewhere on the order of $5,000 with API metered tokens, and I know how much tokens cost because that's where I live" "I ran this overnight multiple times. It's very possible I could have burned through $20,000 worth of API credits for $200. I knew it was subsidized, but it's subsidized to a level that's crazy. It's 100x subsidized"

Laura Shin

85,135 views • 3 months ago

Trump: Saudi Arabia, Kuwait is getting hit, Bahrain is getting—all these countries are getting hit. There was no expert that would say that was going to happen. It's not a question of like gee, should you have known? And if we did, no big deal. I mean, we have to do what we have to do

Trump: Saudi Arabia, Kuwait is getting hit, Bahrain is getting—all these countries are getting hit. There was no expert that would say that was going to happen. It's not a question of like gee, should you have known? And if we did, no big deal. I mean, we have to do what we have to do

Acyn

120,685 views • 6 months ago

California's high-speed rail project would take 216 years to complete at Democrats current pace, which is 3.5 miles per year - Approved in 2008 with a $33 billion taxpayer budget, the 500-mile high-speed rail project was meant to connect LA and San Francisco - After 16 years, only 57 miles are track ready - This is an average of just 3.5 miles per year - At this pace, it would take 141 years to complete phase 1 - It would take 216 years to complete phase 2 - Real estimated cost has now reached over $200 billion dollars - $17 billion has already been spent - As of this video, only 57 miles have been completed This is Governor Gavin Newsom’s project

California's high-speed rail project would take 216 years to complete at Democrats current pace, which is 3.5 miles per year - Approved in 2008 with a $33 billion taxpayer budget, the 500-mile high-speed rail project was meant to connect LA and San Francisco - After 16 years, only 57 miles are track ready - This is an average of just 3.5 miles per year - At this pace, it would take 141 years to complete phase 1 - It would take 216 years to complete phase 2 - Real estimated cost has now reached over $200 billion dollars - $17 billion has already been spent - As of this video, only 57 miles have been completed This is Governor Gavin Newsom’s project

Wall Street Apes

33,170 views • 5 months ago

How to cut your AI bill by 60% switching to Opus 5.5: Most people will swap the model name, save 24%, and stop there. The other 37% is sitting in your settings. Here's the real before and after on an example agent workload. One month: 100M input tokens (80M cached reads, 5M cache writes, 15M uncached) 10M output tokens BEFORE: Opus 5 Cache reads: 80M × $0.50 = $40 Cache writes: 5M × $6.25 = $31.25 Uncached input: 15M × $5 = $75 Output: 10M × $25 = $250 Total: $396.25 STEP 1. Just switch the model. Same tokens, new prices. Cache reads $0.20. Writes $5. Input $4. Output $20. $16 + $25 + $60 + $200 = $301 24% cheaper. That's what everyone's screenshotting. STEP 2. Let it write less. Box measured Opus 5.5 at 40% less verbose with no drop in accuracy. 10M output tokens becomes 6M. Output drops from $200 to $120. Total: $221. Now you're at 44% off. STEP 3. Stop running every call at max effort. Thinking can't be switched off on 5.5 anymore, so effort is your lever. Classifying, routing, formatting, summarizing? Drop the effort. Save the high settings for the calls that actually reason. Say that trims output another 25%, 6M down to 4.5M. Output: $90. Total: $191. 52% off. STEP 4. Cache the stuff you keep resending. This is the one nobody does. Cache reads went from $0.50 to $0.20. That's 60% off the cheapest line on your bill. Move your system prompt, tool definitions and repo context into the cache. Uncached input drops from 15M to 5M, cache reads go up to 90M. $18 + $25 + $20 + $90 = $153 AFTER: $153 From $396.25. 61% cheaper. Same work. The model gave you 24%. You gave yourself the other 37%. Anthropic's own number is 40% cheaper than Opus 5 on typical workloads. Your number depends on how much of your bill is output and how much of your prompt you're resending uncached every call. So check those two first. Bookmark this for when you migrate. follow CyrilXBT

How to cut your AI bill by 60% switching to Opus 5.5: Most people will swap the model name, save 24%, and stop there. The other 37% is sitting in your settings. Here's the real before and after on an example agent workload. One month: 100M input tokens (80M cached reads, 5M cache writes, 15M uncached) 10M output tokens BEFORE: Opus 5 Cache reads: 80M × $0.50 = $40 Cache writes: 5M × $6.25 = $31.25 Uncached input: 15M × $5 = $75 Output: 10M × $25 = $250 Total: $396.25 STEP 1. Just switch the model. Same tokens, new prices. Cache reads $0.20. Writes $5. Input $4. Output $20. $16 + $25 + $60 + $200 = $301 24% cheaper. That's what everyone's screenshotting. STEP 2. Let it write less. Box measured Opus 5.5 at 40% less verbose with no drop in accuracy. 10M output tokens becomes 6M. Output drops from $200 to $120. Total: $221. Now you're at 44% off. STEP 3. Stop running every call at max effort. Thinking can't be switched off on 5.5 anymore, so effort is your lever. Classifying, routing, formatting, summarizing? Drop the effort. Save the high settings for the calls that actually reason. Say that trims output another 25%, 6M down to 4.5M. Output: $90. Total: $191. 52% off. STEP 4. Cache the stuff you keep resending. This is the one nobody does. Cache reads went from $0.50 to $0.20. That's 60% off the cheapest line on your bill. Move your system prompt, tool definitions and repo context into the cache. Uncached input drops from 15M to 5M, cache reads go up to 90M. $18 + $25 + $20 + $90 = $153 AFTER: $153 From $396.25. 61% cheaper. Same work. The model gave you 24%. You gave yourself the other 37%. Anthropic's own number is 40% cheaper than Opus 5 on typical workloads. Your number depends on how much of your bill is output and how much of your prompt you're resending uncached every call. So check those two first. Bookmark this for when you migrate. follow CyrilXBT

CyrilXBT

15,803 views • 3 days ago

A lot of AI agent products promise “automation.” What they actually give you is another dashboard to micromanage. You spend hours wiring prompts, APIs, workflows, memory systems, and tool calls before the agents even begin doing real work. That’s been my experience with most agent stacks so far. Then I tried Helio The difference hit immediately. Instead of configuring everything like an engineer, I just gave the system an outcome. The AI teammates coordinated tasks, delegated work between themselves, and started executing while I was away. No babysitting. No endless tweaking. No orchestration headaches. For the first time, an AI agent product felt less like “using software” and more like hiring a small autonomous team. That’s the direction the market is moving toward fast.

A lot of AI agent products promise “automation.” What they actually give you is another dashboard to micromanage. You spend hours wiring prompts, APIs, workflows, memory systems, and tool calls before the agents even begin doing real work. That’s been my experience with most agent stacks so far. Then I tried Helio The difference hit immediately. Instead of configuring everything like an engineer, I just gave the system an outcome. The AI teammates coordinated tasks, delegated work between themselves, and started executing while I was away. No babysitting. No endless tweaking. No orchestration headaches. For the first time, an AI agent product felt less like “using software” and more like hiring a small autonomous team. That’s the direction the market is moving toward fast.

Atul Kumar

34,960 views • 3 months ago

Interesting approach to long context/continual learning here from Baseten at Cursor Compact long trajectories by compressing a prefix of the KV cache using an MLP/autoencoder. You can train this "compactor" MLP by learning to reconstruct activations that the original KV cache would produce on subsequent tokens. This maximally reconstructs information from long context that's useful for subsequent outputs. If you run an agent over extremely long context and run this compaction recursively, the activations of this compressed KV cache become like trained weights. Similar to task-specific LoRA or "cartridges". Would not be surprised if OpenAI is running a similar algorithm for their blackbox compaction. Clear benefits here if you can avoid busting the cache, as e.g. compacting via writing to a text file would. Seeing many emerging approaches to fixed size latent states for LLMs that seem promising. If building a task-specific KV cache compression ends up being more sample efficient than running backprop, getting this to "work" feels like one of the 2-3 remaining breakthroughs on the path to true AGI

Interesting approach to long context/continual learning here from Baseten at Cursor Compact long trajectories by compressing a prefix of the KV cache using an MLP/autoencoder. You can train this "compactor" MLP by learning to reconstruct activations that the original KV cache would produce on subsequent tokens. This maximally reconstructs information from long context that's useful for subsequent outputs. If you run an agent over extremely long context and run this compaction recursively, the activations of this compressed KV cache become like trained weights. Similar to task-specific LoRA or "cartridges". Would not be surprised if OpenAI is running a similar algorithm for their blackbox compaction. Clear benefits here if you can avoid busting the cache, as e.g. compacting via writing to a text file would. Seeing many emerging approaches to fixed size latent states for LLMs that seem promising. If building a task-specific KV cache compression ends up being more sample efficient than running backprop, getting this to "work" feels like one of the 2-3 remaining breakthroughs on the path to true AGI

Jay Hack

29,792 views • 2 months ago

In 2024, the city of Bangor allocated $250,000 to build 6 public bathrooms. They have built four of them so far at a cost of ~$150k. In under two years, this is what they look like... Just a complete and total disaster:

In 2024, the city of Bangor allocated $250,000 to build 6 public bathrooms. They have built four of them so far at a cost of ~$150k. In under two years, this is what they look like... Just a complete and total disaster:

The Maine Wire

26,653 views • 5 months ago

DeepSeek V4.1 Flash is a very interesting model. I just tested it on the BridgeBench lava lamp test and it took LONGER to complete than Fable 5.1 and GPT 6 Astra. It ran at 344 toks/sec, spent 23.5M tokens with a cache hit rate of 99.7%, and cost $0.33. Even though it runs so fast it ends up completing tasks very slowly. The lava lamp that it produced was better than Muse Spark 1.3 and Gemini 3.8 Flash.

DeepSeek V4.1 Flash is a very interesting model. I just tested it on the BridgeBench lava lamp test and it took LONGER to complete than Fable 5.1 and GPT 6 Astra. It ran at 344 toks/sec, spent 23.5M tokens with a cache hit rate of 99.7%, and cost $0.33. Even though it runs so fast it ends up completing tasks very slowly. The lava lamp that it produced was better than Muse Spark 1.3 and Gemini 3.8 Flash.

BridgeMind

128,202 views • 17 days ago