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.

7K+ RL environments from Xiaomi across coding, cyber, general, music, and more. You can now explore, visualize, and actually run them. Pick an environment → pick a model → run the rollout → watch exactly what happens RL environments are a lot more interesting when you can actually play... show more

Adithya S K

15,647 subscribers

17,667 views • 5 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

6 Comments

Adithya S K5 days ago

This is one of the best resources to actually understand RL environment design. Have fun exploring and drop in any feedback !!

Roni Rechter4 days ago

7000 environments and picking one is still the hard part

Divyendra Singh5 days ago

can you compare two model rollouts side by side?

Adithya S K5 days ago

thats a good idea will add the comparision view

Blue Guy5 days ago

Reward model column with ground truth per prompt is the part worth staring at. Most open RL datasets skip that and just dump rollouts.

Raafi4 days ago

I went through the space yesterday great work @adithya_s_k , just that while checking each task file I found it too compact to read, maybe it can be improved for enhancing readability, great stuff 👏

Related Videos

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

66,691 views • 1 month ago

Our model can now learn from its own experience with RL! Our new π*0.6 model can more than double throughput over a base model trained without RL, and can perform real-world tasks: making espresso drinks, folding diverse laundry, and assembling boxes. More in the thread below.

Our model can now learn from its own experience with RL! Our new π*0.6 model can more than double throughput over a base model trained without RL, and can perform real-world tasks: making espresso drinks, folding diverse laundry, and assembling boxes. More in the thread below.

Physical Intelligence

711,116 views • 10 months ago

🚀 Today we're launching LangSmith Sandboxes Agents get a lot more useful when they can run code: analyze data, call APIs, build entire applications. Sandboxes give them a safe place to do it with ephemeral, locked-down environments you control. Now in Private Preview. Learn more: Join the waitlist:

🚀 Today we're launching LangSmith Sandboxes Agents get a lot more useful when they can run code: analyze data, call APIs, build entire applications. Sandboxes give them a safe place to do it with ephemeral, locked-down environments you control. Now in Private Preview. Learn more: Join the waitlist:

LangChain

46,479 views • 6 months ago

With RL, the robot can learn very precise tasks, like fastening a zip tie, and can actually do it more consistently and more quickly than even human teleoperation.

With RL, the robot can learn very precise tasks, like fastening a zip tie, and can actually do it more consistently and more quickly than even human teleoperation.

Physical Intelligence

19,075 views • 6 months ago

Introducing Reca: A voice companion for Indian parents and grandparents. It talks to them, grows with them, can play games, order groceries, create learning plans and can actually help them pick up a new skill! Watch Reca in action ⬇️

Introducing Reca: A voice companion for Indian parents and grandparents. It talks to them, grows with them, can play games, order groceries, create learning plans and can actually help them pick up a new skill! Watch Reca in action ⬇️

shivansh

10,298 views • 26 days ago

Since everyone is talking about RL Environments and GRPO now but no one knows how it works we thought it would be cool to make an explainer video + code you can run: This is an example of using GRPO to train Qwen 2.5 to play 2048 (code in thread) 🧵:

Since everyone is talking about RL Environments and GRPO now but no one knows how it works we thought it would be cool to make an explainer video + code you can run: This is an example of using GRPO to train Qwen 2.5 to play 2048 (code in thread) 🧵:

Jay

153,216 views • 1 year ago

Hiring RL Engineer! Started off as a curious project at Lossfunk to push the boundaries of LLMs in social reasoning - we are now building RL environments, data, and benchmarks to simulate more real-world scenarios. If you want to train SoTA RL models over multi-GPUs (H200s/B200s) to unlock next AI frontier, this is for you.

Hiring RL Engineer! Started off as a curious project at Lossfunk to push the boundaries of LLMs in social reasoning - we are now building RL environments, data, and benchmarks to simulate more real-world scenarios. If you want to train SoTA RL models over multi-GPUs (H200s/B200s) to unlock next AI frontier, this is for you.

Satpal Singh Rathore

46,010 views • 1 year ago

Introducing INTELLECT-3: Scaling RL to a 100B+ MoE model on our end-to-end stack Achieving state-of-the-art performance for its size across math, code and reasoning Built using the same tools we put in your hands, from environments & evals, RL frameworks, sandboxes & more

Introducing INTELLECT-3: Scaling RL to a 100B+ MoE model on our end-to-end stack Achieving state-of-the-art performance for its size across math, code and reasoning Built using the same tools we put in your hands, from environments & evals, RL frameworks, sandboxes & more

Prime Intellect

1,150,419 views • 10 months ago

With the general availability of Cloud Run jobs and the Cloud Run second-generation execution environment, the sky’s the limit for what you can build on our serverless platforms 🎉 Learn more ↓

With the general availability of Cloud Run jobs and the Cloud Run second-generation execution environment, the sky’s the limit for what you can build on our serverless platforms 🎉 Learn more ↓

Google Cloud Tech

10,949 views • 3 years ago

Starting today, you can run cloud agents inside fully configured development environments. Set them up the same way you'd set up a laptop for an engineer: cloned repos, installed dependencies, and toolchain credentials.

Starting today, you can run cloud agents inside fully configured development environments. Set them up the same way you'd set up a laptop for an engineer: cloned repos, installed dependencies, and toolchain credentials.

Cursor

290,328 views • 4 months ago

Spatial Wallpapers are here! I just got my first hands-on experience with the new Spatial Panoramas and Custom Environments features, and it's incredible how immersive these memories feel when they surround you with real depth. With just a few taps, you can save them to a new folder in the Environments view and use them as your own custom spatial wallpapers.

Spatial Wallpapers are here! I just got my first hands-on experience with the new Spatial Panoramas and Custom Environments features, and it's incredible how immersive these memories feel when they surround you with real depth. With just a few taps, you can save them to a new folder in the Environments view and use them as your own custom spatial wallpapers.

Phil Traut ᯅ

11,202 views • 3 months ago

Women are now hitting 40s and reality slapping them harder. Nature is cruel. The wall is unforgiving and let me tell you Maina, they are now going back to pick 22-year-old Chihuahuas, just for society to accept them. Missing a train is more painful when you run after it.

Sensitive content

Women are now hitting 40s and reality slapping them harder. Nature is cruel. The wall is unforgiving and let me tell you Maina, they are now going back to pick 22-year-old Chihuahuas, just for society to accept them. Missing a train is more painful when you run after it.

Ja Loka

72,068 views • 2 months ago

This is the Claude Code moment for Reinforcement Learning. Every frontier lab knows the secret: post-training is where the magic happens. If you can eval a task, you can use RL to benchmax your model. But building and scaling the RL loop has remained a dark art reserved for the top 1% of AI Labs. Until today. Introducing Optimus — an autonomous RL Agent Harness, and the first of many we are about to ship Iacon Autonomics. We are removing the manual grind of RL and adding taste and relentlessness to the process. Just hand Optimus a codebase, an API key, and a goal. It spins up a fleet of sub-agents to run hundreds of parallel experiments. It learns from its own history, autonomously building an evolutionary tree of runs where every failure sets the next batch up to win. Train LLMs, agents, robots in sim, game npcs and much more. If you can score it, Optimus can train it. Stop doing manual RL. Let the agents cook.

This is the Claude Code moment for Reinforcement Learning. Every frontier lab knows the secret: post-training is where the magic happens. If you can eval a task, you can use RL to benchmax your model. But building and scaling the RL loop has remained a dark art reserved for the top 1% of AI Labs. Until today. Introducing Optimus — an autonomous RL Agent Harness, and the first of many we are about to ship Iacon Autonomics. We are removing the manual grind of RL and adding taste and relentlessness to the process. Just hand Optimus a codebase, an API key, and a goal. It spins up a fleet of sub-agents to run hundreds of parallel experiments. It learns from its own history, autonomously building an evolutionary tree of runs where every failure sets the next batch up to win. Train LLMs, agents, robots in sim, game npcs and much more. If you can score it, Optimus can train it. Stop doing manual RL. Let the agents cook.

Dev Shah

13,730 views • 1 month ago

JUST IN: JD Vance trolls "Coach" Tim Walz after the governor claimed you can "run a mean pick 6" in football. "AOC can run a mean pick 6 - and I can call an audible on a play," Walz's X account said in an effort to impress young men. "You don't run a pick-six, you run the West Coast offense or the spread offense," Vance said at a rally. "It made me realize that even though they say Tim Walz was a football coach, I think I know more about football than Tim Walz." "And it's appropriate 'cause I think my running mate, Donald J. Trump knows more about working at McDonald's than Kamala Harris does."

JUST IN: JD Vance trolls "Coach" Tim Walz after the governor claimed you can "run a mean pick 6" in football. "AOC can run a mean pick 6 - and I can call an audible on a play," Walz's X account said in an effort to impress young men. "You don't run a pick-six, you run the West Coast offense or the spread offense," Vance said at a rally. "It made me realize that even though they say Tim Walz was a football coach, I think I know more about football than Tim Walz." "And it's appropriate 'cause I think my running mate, Donald J. Trump knows more about working at McDonald's than Kamala Harris does."

Collin Rugg

4,777,367 views • 1 year ago

With the launch of GLM 5.2 this week, I see everyone asking "have open models caught up to closed models?" The more interesting question that's getting missed: what can you do with an open model that you can't do with a closed one? You can specialize them. And when you do, the number of economically valuable tasks open models can do actually subsumes that of closed models. Charlie O'Neill explaining this:

With the launch of GLM 5.2 this week, I see everyone asking "have open models caught up to closed models?" The more interesting question that's getting missed: what can you do with an open model that you can't do with a closed one? You can specialize them. And when you do, the number of economically valuable tasks open models can do actually subsumes that of closed models. Charlie O'Neill explaining this:

Madison Kanna

18,149 views • 3 months ago

Using managed agents is perfect for running coding agents anywhere in the world. The beauty of this: • You can resume a session from anywhere • You can fork a session into multiple agents • You can run parallel agents • You can reuse environments Here is how you can do this with Claude Code, Codex, OpenCode, or Hermes.

Using managed agents is perfect for running coding agents anywhere in the world. The beauty of this: • You can resume a session from anywhere • You can fork a session into multiple agents • You can run parallel agents • You can reuse environments Here is how you can do this with Claude Code, Codex, OpenCode, or Hermes.

Santiago

15,931 views • 8 days ago

You can now run Claude Code tasks from the Claude Android app, in research preview. Kick off cloud-based tasks from your phone, let Claude run, then pick them up later to review and merge work. Download Claude for Android:

You can now run Claude Code tasks from the Claude Android app, in research preview. Kick off cloud-based tasks from your phone, let Claude run, then pick them up later to review and merge work. Download Claude for Android:

Claude

51,409 views • 9 months ago

#PixVerseWorldModel is doing something pretty interesting with AI video. One of the things that stood out to me is the WASD experience. AI video usually follows a simple workflow: write a prompt, generate a clip, and watch the result. With PixVerse World Model, you can actually move around inside the generated world, control the camera, and use prompts to influence the environment, characters, and actions as you move through the scene. The interesting part is how movement and prompting work together. Instead of generating a completely new clip every time you want something to change, you can keep interacting with the world and see how it responds. Text prompts start to feel less like instructions for generating a video and more like another way of interacting with the world itself. It makes PixVerse World Model feel less like traditional AI video and more like a playable, generated environment. Maybe the more interesting question for AI video isn’t just “what can it generate?” It’s “what can we actually do once we’re inside it?” Try now:

#PixVerseWorldModel is doing something pretty interesting with AI video. One of the things that stood out to me is the WASD experience. AI video usually follows a simple workflow: write a prompt, generate a clip, and watch the result. With PixVerse World Model, you can actually move around inside the generated world, control the camera, and use prompts to influence the environment, characters, and actions as you move through the scene. The interesting part is how movement and prompting work together. Instead of generating a completely new clip every time you want something to change, you can keep interacting with the world and see how it responds. Text prompts start to feel less like instructions for generating a video and more like another way of interacting with the world itself. It makes PixVerse World Model feel less like traditional AI video and more like a playable, generated environment. Maybe the more interesting question for AI video isn’t just “what can it generate?” It’s “what can we actually do once we’re inside it?” Try now:

Nayeem Sheikh

14,959 views • 10 days ago

pick a city. pick a year. see what a property bought then would be worth now. the mey time machine covers 22 of them, from shanghai in 2000 to hanoi today. run it once, and you won't be able to stop →

pick a city. pick a year. see what a property bought then would be worth now. the mey time machine covers 22 of them, from shanghai in 2000 to hanoi today. run it once, and you won't be able to stop →

mey real

19,634 views • 2 months ago

This is what happens when you take a skull merchant player who knows all the nooks and crannies a drone can scan you through, and give them a Killer with an actually strong power that does stuff. I love Jason SO FUCKING MUCH. This is everything I've wanted from a ranged Killer.

This is what happens when you take a skull merchant player who knows all the nooks and crannies a drone can scan you through, and give them a Killer with an actually strong power that does stuff. I love Jason SO FUCKING MUCH. This is everything I've wanted from a ranged Killer.

Pixel Bush

34,060 views • 4 months ago