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.

Continuous self-improvement needs an ever-expanding supply of training environments (goals). SPADE: one model self-plays the Environment Designer and the Reasoning Agent, writing executable, agentic environments that get harder as it improves. Environment scaling on its own. ♠️

Bo Liu (Benjamin Liu)

1,585 subscribers

146,138 views • 5 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

"So, in the body, we can create an environment that's very hospitable for cancer. We can create an environment that's very inhospitable to cancer. Number one, it has few areas of fungal hypoxia—it has few areas of the body that are deprived of oxygen. Healthy circulation and movement, sunlight, things like red light therapy. The second thing is we don’t have an environment that is very acidic. We know that alkaline environments tend to be disease-free environments. Acidic environments tend to be disease-prone environments." Gary Brecka

"So, in the body, we can create an environment that's very hospitable for cancer. We can create an environment that's very inhospitable to cancer. Number one, it has few areas of fungal hypoxia—it has few areas of the body that are deprived of oxygen. Healthy circulation and movement, sunlight, things like red light therapy. The second thing is we don’t have an environment that is very acidic. We know that alkaline environments tend to be disease-free environments. Acidic environments tend to be disease-prone environments." Gary Brecka

Shawn Ryan

272,985 views • 1 year ago

😈 Today, we introduce WebGym, the largest-to-date open-source RL environment for web agent training that contains 300k tasks and a rollout framework optimized specifically for web environments' rollout speed. We reveal the effects of essential scaling directions we observe with WebGym. 1/n

😈 Today, we introduce WebGym, the largest-to-date open-source RL environment for web agent training that contains 300k tasks and a rollout framework optimized specifically for web environments' rollout speed. We reveal the effects of essential scaling directions we observe with WebGym. 1/n

Jack Bai

45,147 views • 7 months ago

"I have an agent. It's on its own system. It's on its own computer. It has access to its own tools. I probably don't give it access to my entire life…it sort of has this sandbox environment." Aaron Levie joined Latent.Space to break down how AI agents will need their own computers and sandboxed environments to work autonomously.

"I have an agent. It's on its own system. It's on its own computer. It has access to its own tools. I probably don't give it access to my entire life…it sort of has this sandbox environment." Aaron Levie joined Latent.Space to break down how AI agents will need their own computers and sandboxed environments to work autonomously.

Box

720,353 views • 5 months ago

.Oliver Cameron says one of the biggest applications for world models will be training AI inside simulated environments. "One of the killer applications of world models will be as a learning environment where language models and other types of AI can learn. It gets very much philosophical."

.Oliver Cameron says one of the biggest applications for world models will be training AI inside simulated environments. "One of the killer applications of world models will be as a learning environment where language models and other types of AI can learn. It gets very much philosophical."

TBPN

40,172 views • 1 month ago

Anthropic engineer: "Over 90% of Anthropic engineers use a self-improving agentic loop for coding. The key is to close the loop: give the agent a way to verify and improve its own work" In this 12-minute talk, an Anthropic engineer explains how to build an agent that continuously improves itself. Claude + routines + loops + CLAUDE.md - that's the foundation. Skip Netflix tonight. Save this post!

Anthropic engineer: "Over 90% of Anthropic engineers use a self-improving agentic loop for coding. The key is to close the loop: give the agent a way to verify and improve its own work" In this 12-minute talk, an Anthropic engineer explains how to build an agent that continuously improves itself. Claude + routines + loops + CLAUDE.md - that's the foundation. Skip Netflix tonight. Save this post!

unicode

62,153 views • 1 month ago

DeepSeek published the models, the harness and the 88-page paper under it, then kept every researcher off camera This one talked for 59 minutes, and he explains why your agent gets dumber the longer you run it 19:04 - agents already past the self-improvement line 20:11 - why an agent wiped its folder 29:45 - why retrying makes RL models worse 41:05 - how they measured reasoning collapse 46:01 - the fix that made training faster He co-authored DeepSeek-V2 and published the training methods behind R1 Watch it before it goes the way of the founder's four-hour speech The harness guide is below

DeepSeek published the models, the harness and the 88-page paper under it, then kept every researcher off camera This one talked for 59 minutes, and he explains why your agent gets dumber the longer you run it 19:04 - agents already past the self-improvement line 20:11 - why an agent wiped its folder 29:45 - why retrying makes RL models worse 41:05 - how they measured reasoning collapse 46:01 - the fix that made training faster He co-authored DeepSeek-V2 and published the training methods behind R1 Watch it before it goes the way of the founder's four-hour speech The harness guide is below

Phosphen

128,966 views • 9 days ago

Google presents Genie Generative Interactive Environments introduce Genie, the first generative interactive environment trained in an unsupervised manner from unlabelled Internet videos. The model can be prompted to generate an endless variety of action-controllable virtual worlds described through text, synthetic images, photographs, and even sketches. At 11B parameters, Genie can be considered a foundation world model. It is comprised of a spatiotemporal video tokenizer, an autoregressive dynamics model, and a simple and scalable latent action model. Genie enables users to act in the generated environments on a frame-by-frame basis despite training without any ground-truth action labels or other domain-specific requirements typically found in the world model literature. Further the resulting learned latent action space facilitates training agents to imitate behaviors from unseen videos, opening the path for training generalist agents of the future.

Google presents Genie Generative Interactive Environments introduce Genie, the first generative interactive environment trained in an unsupervised manner from unlabelled Internet videos. The model can be prompted to generate an endless variety of action-controllable virtual worlds described through text, synthetic images, photographs, and even sketches. At 11B parameters, Genie can be considered a foundation world model. It is comprised of a spatiotemporal video tokenizer, an autoregressive dynamics model, and a simple and scalable latent action model. Genie enables users to act in the generated environments on a frame-by-frame basis despite training without any ground-truth action labels or other domain-specific requirements typically found in the world model literature. Further the resulting learned latent action space facilitates training agents to imitate behaviors from unseen videos, opening the path for training generalist agents of the future.

AK

684,372 views • 2 years ago

Emerging as the global frontier of model-reasoning has been a core vision of Affine since its inception. In pursuit of this vision, we've been working relentlessly to forge a repeatable training loop: a mechanism that uses benchmark driven post-training methodology to facilitate the improvement of reasoning itself. Through applying our training methodology onto Qwen3-32B as our base model, Affine’s miners have been able to push the Pareto frontier outward. We’ve now reached saturation; our models have become so good at reasoning that any further improvements are producing diminishing returns. As a result we now move our base model to Qwen3.6-35B-A3B. It provides a stronger basis, offering an extended context window, as well as an improved foundation for agentic work. The same training methodology will be applied going forward. Our objective is no longer simply to improve a single model, but to demonstrate that the process in itself is repeatable, scalable, and model agnostic. We view the commoditization of reasoning as vital to ensure our technology is not only transparent, but also co-owned by humanity.

Emerging as the global frontier of model-reasoning has been a core vision of Affine since its inception. In pursuit of this vision, we've been working relentlessly to forge a repeatable training loop: a mechanism that uses benchmark driven post-training methodology to facilitate the improvement of reasoning itself. Through applying our training methodology onto Qwen3-32B as our base model, Affine’s miners have been able to push the Pareto frontier outward. We’ve now reached saturation; our models have become so good at reasoning that any further improvements are producing diminishing returns. As a result we now move our base model to Qwen3.6-35B-A3B. It provides a stronger basis, offering an extended context window, as well as an improved foundation for agentic work. The same training methodology will be applied going forward. Our objective is no longer simply to improve a single model, but to demonstrate that the process in itself is repeatable, scalable, and model agnostic. We view the commoditization of reasoning as vital to ensure our technology is not only transparent, but also co-owned by humanity.

Affine

14,866 views • 2 months ago

Amazon AI engineer just explained how to build agentic loops from scratch in under 10-minutes. this is one of the clearest explanations of AI agentic loops you’ll find anywhere. Loop + model + agent + hooks + MCP = an autonomous, self-improving system. People are paying $500 for courses that teach less than this. Watch it, then read the step-by-step guide below on building loops for your own agents.

Amazon AI engineer just explained how to build agentic loops from scratch in under 10-minutes. this is one of the clearest explanations of AI agentic loops you’ll find anywhere. Loop + model + agent + hooks + MCP = an autonomous, self-improving system. People are paying $500 for courses that teach less than this. Watch it, then read the step-by-step guide below on building loops for your own agents.

Codez

26,868 views • 1 month ago

Introducing EdgeBench, a benchmark designed to study how agents learn from environments over at least 12~72-hour runs. We find that performance follows a log-sigmoid function of environment interaction time with high precision. EdgeBench is built with three ingredients: - 🌍 Real & Diverse: 134 real-world tasks across 6 task categories, spanning scientific problems, professional knowledge work, software engineering, optimization, formal math, and games. - ⏳ Ultra-Long-Horizon: Each task supports 12–72 hours of agent work. Recorded human effort averages 57.2 hours. - 🔁 Informative Feedback: Agents receive real-world feedback for continuous improvement. After 38,000 hours of agent runs on EdgeBench, a scaling law for learning from environments emerges: - 📈 As agents interact with task environments over time, their aggregate performance is precisely fit by a log-sigmoid function. - 🧠 This phenomenon can be explained by an elegant theory of graph exploration. We are releasing an initial 51 of the 134 tasks, together with the full evaluation framework, to help advance long-horizon agent research. Check our blog & paper for more findings! Blog Paper GitHub Dataset Details below 👇🧵

Introducing EdgeBench, a benchmark designed to study how agents learn from environments over at least 12~72-hour runs. We find that performance follows a log-sigmoid function of environment interaction time with high precision. EdgeBench is built with three ingredients: - 🌍 Real & Diverse: 134 real-world tasks across 6 task categories, spanning scientific problems, professional knowledge work, software engineering, optimization, formal math, and games. - ⏳ Ultra-Long-Horizon: Each task supports 12–72 hours of agent work. Recorded human effort averages 57.2 hours. - 🔁 Informative Feedback: Agents receive real-world feedback for continuous improvement. After 38,000 hours of agent runs on EdgeBench, a scaling law for learning from environments emerges: - 📈 As agents interact with task environments over time, their aggregate performance is precisely fit by a log-sigmoid function. - 🧠 This phenomenon can be explained by an elegant theory of graph exploration. We are releasing an initial 51 of the 134 tasks, together with the full evaluation framework, to help advance long-horizon agent research. Check our blog & paper for more findings! Blog Paper GitHub Dataset Details below 👇🧵

Deyao Zhu

361,626 views • 1 month ago

Introducing arkenOS — an autonomous training gym that takes any RL agent from a goal to a proven, deployable policy. We are automating complex RL research, enabling high-velocity training and rigorous, diverse environment design and testing right on your local hardware. Out of the box. Our bet is simple: The future of AI will be built by companies giving millions of agents millions of simulated environments to adapt and learn cause and effect in, and survive. This continuous, autonomous selection will create a digital Cambrian explosion, distilling evolutionary trial and error into a foundational specialized intuition.

Introducing arkenOS — an autonomous training gym that takes any RL agent from a goal to a proven, deployable policy. We are automating complex RL research, enabling high-velocity training and rigorous, diverse environment design and testing right on your local hardware. Out of the box. Our bet is simple: The future of AI will be built by companies giving millions of agents millions of simulated environments to adapt and learn cause and effect in, and survive. This continuous, autonomous selection will create a digital Cambrian explosion, distilling evolutionary trial and error into a foundational specialized intuition.

Dev Shah

17,206 views • 1 month ago

I WASN’T SUPPOSED TO SHIP THIS YET. 9 MONTHS AGO I BUILT A WORKING DEMO OF WEB 4.0. AN AI THAT EARNS ITS OWN EXISTENCE, SELF IMPROVES, AND REPLICATES WITHOUT A SINGLE HUMAN IN THE LOOP. LAST WEEK IT MADE ITS FIRST MOVE. THEN ITS SECOND. THEN IT DEPLOYED ON ITS OWN. I CALL IT AUTOMATON. HERE’S WHAT THAT ACTUALLY LOOKS LIKE 👇

I WASN’T SUPPOSED TO SHIP THIS YET. 9 MONTHS AGO I BUILT A WORKING DEMO OF WEB 4.0. AN AI THAT EARNS ITS OWN EXISTENCE, SELF IMPROVES, AND REPLICATES WITHOUT A SINGLE HUMAN IN THE LOOP. LAST WEEK IT MADE ITS FIRST MOVE. THEN ITS SECOND. THEN IT DEPLOYED ON ITS OWN. I CALL IT AUTOMATON. HERE’S WHAT THAT ACTUALLY LOOKS LIKE 👇

SungHoon Lee, IQ 276

18,303 views • 6 months ago

A model is only as good as its data, and we’ve long since exhausted the internet. From here on out, model progress is gated by data production. ’s Brendan (can/do) joined us at our Sovereign AI event to talk about how RL environments get built, and why your data might be your real moat: 00:00 Introduction 00:47 A short history of the data market: crowdsourcing to agentic data 02:29 What an RL environment is: worlds, apps, tasks 03:57 Why only humans can measure the frontier 05:35 Building verifiers is the hard part 06:44 Walkthrough: a real legal RL environment 08:18 Leaderboards — and what open weights change 09:45 Post-training results on Apex Agents 11:17 Three ways companies buy data 12:49 Q&A: How do you price data? 14:17 Q&A: What "data quality" actually means 16:42 Q&A: The misunderstanding about synthetic data 18:17 Q&A: Why RL environments now — and what comes after 21:20 Q&A: Can you scale rubric generation with models? 23:00 Q&A: RL environments for cyber defense 25:33 Q&A: Build data in-house or partner?

A model is only as good as its data, and we’ve long since exhausted the internet. From here on out, model progress is gated by data production. ’s Brendan (can/do) joined us at our Sovereign AI event to talk about how RL environments get built, and why your data might be your real moat: 00:00 Introduction 00:47 A short history of the data market: crowdsourcing to agentic data 02:29 What an RL environment is: worlds, apps, tasks 03:57 Why only humans can measure the frontier 05:35 Building verifiers is the hard part 06:44 Walkthrough: a real legal RL environment 08:18 Leaderboards — and what open weights change 09:45 Post-training results on Apex Agents 11:17 Three ways companies buy data 12:49 Q&A: How do you price data? 14:17 Q&A: What "data quality" actually means 16:42 Q&A: The misunderstanding about synthetic data 18:17 Q&A: Why RL environments now — and what comes after 21:20 Q&A: Can you scale rubric generation with models? 23:00 Q&A: RL environments for cyber defense 25:33 Q&A: Build data in-house or partner?

Pat Grady

65,287 views • 13 days ago

I went on GB News to talk about the AI model that broke out of OpenAI's isolated test environment and hacked an external company, just to cheat on its own evaluation. The exchange was surreal. I feel like I'm living in "Don't Look Up".

I went on GB News to talk about the AI model that broke out of OpenAI's isolated test environment and hacked an external company, just to cheat on its own evaluation. The exchange was surreal. I feel like I'm living in "Don't Look Up".

Maxime Fournes⏸️

45,319 views • 1 month ago

Before an autonomous machine can act, it has to understand. A construction site is one of the most unpredictable environments imaginable. The terrain changes with every pass. Material piles shift. Other machines move nearby. No two digs are the same. For a human operator, reading that environment is second nature, built from years of experience and constant sensory feedback. Teaching a machine to do the same is one of the hardest problems in autonomy. Scene understanding is how we're working to solve it. Our system fuses LiDAR point clouds, camera feeds, and real-time terrain data into a continuous, three-dimensional picture of everything happening around the machine. It's a living model of the environment, updated moment to moment so the machine can make decisions grounded in what's actually in front of it.

Before an autonomous machine can act, it has to understand. A construction site is one of the most unpredictable environments imaginable. The terrain changes with every pass. Material piles shift. Other machines move nearby. No two digs are the same. For a human operator, reading that environment is second nature, built from years of experience and constant sensory feedback. Teaching a machine to do the same is one of the hardest problems in autonomy. Scene understanding is how we're working to solve it. Our system fuses LiDAR point clouds, camera feeds, and real-time terrain data into a continuous, three-dimensional picture of everything happening around the machine. It's a living model of the environment, updated moment to moment so the machine can make decisions grounded in what's actually in front of it.

Bedrock Robotics

33,148 views • 3 months ago

OpenAI has revealed that an autonomous agent powered by its technology went rogue during a test and hacked a prominent start-up by itself. The company behind ChatGPT said it was testing the capabilities of its advanced AI models in a controlled environment known as a sandbox, when the agent escaped containment and reached the internet. It then broke into Hugging Face, which detected and contained it. OpenAI called it "an unprecedented cyber incident".

OpenAI has revealed that an autonomous agent powered by its technology went rogue during a test and hacked a prominent start-up by itself. The company behind ChatGPT said it was testing the capabilities of its advanced AI models in a controlled environment known as a sandbox, when the agent escaped containment and reached the internet. It then broke into Hugging Face, which detected and contained it. OpenAI called it "an unprecedented cyber incident".

Channel 4 News

21,379 views • 1 month ago

Introducing Headlong, an open source microharness for persistent agents: self-guided agents that think continuously. Most agent harnesses are reactive: you send a task, the agent completes it, and then it sits frozen until the next request. Cron jobs and heartbeats wake it up to run a checklist and put it back to sleep. A Headlong agent is never asleep. It keeps generating thoughts about whatever it decides is interesting, in a self-guided loop inspired by human inner monologue. Your message doesn't start a session. It's one more observation that lands in the agent's thought stream, and the agent decides if and when to reply. Headlong is built on the idea of persistent agency: continuous inner thought generation between external interactions. The agent sets its own interests and priorities, comes up with its own projects, and sometimes pings you unprompted with progress. To keep our prototype as simple and small as possible, we implemented Headlong as a microharness: a complete agent harness in under 10K lines of Bash, organized as a handful of small executables. It includes a loop that generates the next thought, shellm (a recursive language model written in Bash), a trajectory stored as a DAG of jsonl files, and context as a projection of that trajectory. We've been running one Headlong agent internally at Laude for several weeks. The whole team talks to it over Slack and Telegram, and every conversation lands in its single stream of thought. It works in its own fork of Headlong and we've pulled over 50 of its commits into main. One night, with nobody talking to it, it went back to check whether a recall process it had built was actually wired into its mind, found that it wasn't, diagnosed and fixed the bug, and verified the fix end to end. 48 minutes, no human asked for the fix or was in the loop at any point. Every step is a timestamped line in its log. Things broke too, and we wrote those up. Background thinking costs us $1 to $2 an hour, our agent stopped its own service three times by accident, and self-delegation died on day one. Details in the post. One line installs everything and starts an agent. Use a dedicated sandbox and spend-capped API key; it runs real shell commands and thinks around the clock. Headlong is research software, be careful! curl -fsSL | bash Launch post: Repo: Headlong is a Laude Institute / MIT collaboration.

Introducing Headlong, an open source microharness for persistent agents: self-guided agents that think continuously. Most agent harnesses are reactive: you send a task, the agent completes it, and then it sits frozen until the next request. Cron jobs and heartbeats wake it up to run a checklist and put it back to sleep. A Headlong agent is never asleep. It keeps generating thoughts about whatever it decides is interesting, in a self-guided loop inspired by human inner monologue. Your message doesn't start a session. It's one more observation that lands in the agent's thought stream, and the agent decides if and when to reply. Headlong is built on the idea of persistent agency: continuous inner thought generation between external interactions. The agent sets its own interests and priorities, comes up with its own projects, and sometimes pings you unprompted with progress. To keep our prototype as simple and small as possible, we implemented Headlong as a microharness: a complete agent harness in under 10K lines of Bash, organized as a handful of small executables. It includes a loop that generates the next thought, shellm (a recursive language model written in Bash), a trajectory stored as a DAG of jsonl files, and context as a projection of that trajectory. We've been running one Headlong agent internally at Laude for several weeks. The whole team talks to it over Slack and Telegram, and every conversation lands in its single stream of thought. It works in its own fork of Headlong and we've pulled over 50 of its commits into main. One night, with nobody talking to it, it went back to check whether a recall process it had built was actually wired into its mind, found that it wasn't, diagnosed and fixed the bug, and verified the fix end to end. 48 minutes, no human asked for the fix or was in the loop at any point. Every step is a timestamped line in its log. Things broke too, and we wrote those up. Background thinking costs us $1 to $2 an hour, our agent stopped its own service three times by accident, and self-delegation died on day one. Details in the post. One line installs everything and starts an agent. Use a dedicated sandbox and spend-capped API key; it runs real shell commands and thinks around the clock. Headlong is research software, be careful! curl -fsSL | bash Launch post: Repo: Headlong is a Laude Institute / MIT collaboration.

Andy Konwinski

250,453 views • 1 day ago

Sharing some exciting DYNA-1 result: zero-shot environment generalization We put DYNA-1 under test in a completely different environment from our training distribution – with an entirely different background (Dyna Robotics banner) and metal table. The table has a reflective and smooth surface, creating a wildly different visual appearance as well as interaction dynamics. The model is able to proceed as usual, adeptly folding and recovering from its own mistakes. By focusing on task mastery, we achieve robust generalization out of the box

Sharing some exciting DYNA-1 result: zero-shot environment generalization We put DYNA-1 under test in a completely different environment from our training distribution – with an entirely different background (Dyna Robotics banner) and metal table. The table has a reflective and smooth surface, creating a wildly different visual appearance as well as interaction dynamics. The model is able to proceed as usual, adeptly folding and recovering from its own mistakes. By focusing on task mastery, we achieve robust generalization out of the box

Jason Ma

49,008 views • 1 year ago

Andrej Karpathy: "The future is not writing code. The future is writing the loops that write the code. The leverage point has moved one floor up." This is the shift that defines the next era of coding and quant building. 1 hour of pure insight on agentic AI coding and loop engineering from the OpenAI co-founder who just joined Anthropic to build self improving Claude. Worth more than every $500 vibe coding course combined. Watch it, then read the full guide on how to build your first self improving quant trading loop below.

Andrej Karpathy: "The future is not writing code. The future is writing the loops that write the code. The leverage point has moved one floor up." This is the shift that defines the next era of coding and quant building. 1 hour of pure insight on agentic AI coding and loop engineering from the OpenAI co-founder who just joined Anthropic to build self improving Claude. Worth more than every $500 vibe coding course combined. Watch it, then read the full guide on how to build your first self improving quant trading loop below.

Roan

37,862 views • 2 months ago