正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Was GPT-6 Astra trained on robot data? Got me curious. Borrowed Jay Chooi's code, ran an open-loop rollout on a MolmoAct2 dataset trajectory to find out. Here's what came out 👇

Jiafei Duan

7,230 subscribers

65,163 次观看 • 3 天前 •via X (Twitter)

教育科学技术

Anya Rossi• Live Now

Private livecam show

8 条评论

atharva ☆ 的头像

atharva ☆3 天前

@chooi_jeq exciting

Yitong chen 的头像

Yitong chen3 天前

@chooi_jeq even the order in which things are collected is the same?

Yug 的头像

Yug3 天前

@chooi_jeq Crazy

Bidhan Roy 的头像

Bidhan Roy3 天前

@chooi_jeq nice test. open loop rollout on a known trajectory is the cleanest way to check for memorization

Jiafei Duan 的头像

Jiafei Duan3 天前

@chooi_jeq Only way to trace close source model training data.

Y11 的头像

Y113 天前

@chooi_jeq 实验室玩具，都不是真实世界 ...

Rohan Sikand 🏔️ 的头像

Rohan Sikand 🏔️3 天前

@chooi_jeq what sim is this?

Jiafei Duan 的头像

Jiafei Duan3 天前

@chooi_jeq MolmoAct2 sim

相关视频

we RL trained a 4B VLM to play GeoGuesser > environment, dataset, training recipe, evals, and code are all open-source. > runs on a single A100 > beat gpt 5.4 mini, haiku, qwen 3.5 122B and came close to sonnet on evals

we RL trained a 4B VLM to play GeoGuesser > environment, dataset, training recipe, evals, and code are all open-source. > runs on a single A100 > beat gpt 5.4 mini, haiku, qwen 3.5 122B and came close to sonnet on evals

Adithya S K

67,621 次观看 • 6 天前

GPT-6 Astra is the strongest model for 3D game from a single prompt. We described the game in one sentence > GPT-6 Astra wrote the whole thing - drift physics, combo scoring, near-miss bonuses, speed traps, nitro > out came Street Heat, a full arcade racer running in the browser. The whole pipeline ran end-to-end on Higgsfield Supercomputer.

GPT-6 Astra is the strongest model for 3D game from a single prompt. We described the game in one sentence > GPT-6 Astra wrote the whole thing - drift physics, combo scoring, near-miss bonuses, speed traps, nitro > out came Street Heat, a full arcade racer running in the browser. The whole pipeline ran end-to-end on Higgsfield Supercomputer.

Higgsfield AI 🧩

174,609 次观看 • 11 天前

✨ Introducing 𝐎𝐩𝐞𝐧𝐕𝐋𝐀 — an open-source vision-language-action model for robotics! 👐 - SOTA generalist policy - 7B params - outperforms Octo, RT-2-X on zero-shot evals 🦾 - trained on 970k episodes from OpenX dataset 🤖 - fully open: model/code/data all online 🤗 🧵👇

✨ Introducing 𝐎𝐩𝐞𝐧𝐕𝐋𝐀 — an open-source vision-language-action model for robotics! 👐 - SOTA generalist policy - 7B params - outperforms Octo, RT-2-X on zero-shot evals 🦾 - trained on 970k episodes from OpenX dataset 🤖 - fully open: model/code/data all online 🤗 🧵👇

Moo Jin Kim

227,128 次观看 • 2 年前

GPT-6 Astra is scary good at rebuilding landmarks. This time, the Sydney Opera House. The whole building came out of Astra as code. Blender gave it clean, editable geometry, Cycles gave it path-traced light, all on Higgsfield Supercomputer.

GPT-6 Astra is scary good at rebuilding landmarks. This time, the Sydney Opera House. The whole building came out of Astra as code. Blender gave it clean, editable geometry, Cycles gave it path-traced light, all on Higgsfield Supercomputer.

Higgsfield AI 🧩

115,728 次观看 • 10 天前

I got @openai's GPT-6 astra to edit this video for me with CapCut Transcription here works completely on-device with custom formatting applied using the Google Gemma E4B model's QAT Super happy with what I managed to get out!

I got @openai's GPT-6 astra to edit this video for me with CapCut Transcription here works completely on-device with custom formatting applied using the Google Gemma E4B model's QAT Super happy with what I managed to get out!

Ivan Leo

10,196 次观看 • 10 天前

We are officially in the AGI era🤯 I built this in 2 hours with GPT-6 Astra. > pointed Astra at an open anatomy dataset > Codex wrote the whole interface, all 206 bones > shipped it as a Claude Code Artifact Spin it, blow it apart, cut it in half, click any bone to read what it is. I know nothing about anatomy.

We are officially in the AGI era🤯 I built this in 2 hours with GPT-6 Astra. > pointed Astra at an open anatomy dataset > Codex wrote the whole interface, all 206 bones > shipped it as a Claude Code Artifact Spin it, blow it apart, cut it in half, click any bone to read what it is. I know nothing about anatomy.

Emm | scenario.com

24,606 次观看 • 5 天前

1/ Introducing Glider - the smallest model to beat GPT-4o-mini on eval tasks ⚡🚀 - Open source, open weights, open code - Explainable evaluations by nature - Trained on 183 criteria and 685 domains Try it out for free at 🔥

1/ Introducing Glider - the smallest model to beat GPT-4o-mini on eval tasks ⚡🚀 - Open source, open weights, open code - Explainable evaluations by nature - Trained on 183 criteria and 685 domains Try it out for free at 🔥

PatronusAI

14,856 次观看 • 1 年前

GPT-6 Astra could take the next step in medicine "A full interactive skeleton with all 206 bones got built in two hours" ∙ Astra was pointed at an open anatomy dataset ∙ Codex wrote the entire interface, bone by bone, all 206 of them ∙ It shipped as a Claude Code artifact, ready to open and use The model spins, breaks apart into an exploded view, splits in half, and lets anyone click a single bone to read exactly what it is. Nobody involved in building it had any background in anatomy. That's less a statement about anatomy and more about what two hours with the right tools can now produce.

GPT-6 Astra could take the next step in medicine "A full interactive skeleton with all 206 bones got built in two hours" ∙ Astra was pointed at an open anatomy dataset ∙ Codex wrote the entire interface, bone by bone, all 206 of them ∙ It shipped as a Claude Code artifact, ready to open and use The model spins, breaks apart into an exploded view, splits in half, and lets anyone click a single bone to read exactly what it is. Nobody involved in building it had any background in anatomy. That's less a statement about anatomy and more about what two hours with the right tools can now produce.

Zentrix⌚️

20,901 次观看 • 5 天前

GPT-6 Astra × H3 Max Turbo I asked Astra to build a moving portrait out of hundreds of tiny movies Turbo generated 256 clips at 480p on fal. Astra wrote the pipeline that arranged them into 2,304 moving tiles

GPT-6 Astra × H3 Max Turbo I asked Astra to build a moving portrait out of hundreds of tiny movies Turbo generated 256 clips at 480p on fal. Astra wrote the pipeline that arranged them into 2,304 moving tiles

gokaygokay

28,368 次观看 • 6 天前

timelapse {#33} - started working with convolutional neural networks - build a custom cnn pipeline on mnist dataset - was using bhindiai all along to explain me the code - speed ran the theory part, focus was main on the implementation

timelapse {#33} - started working with convolutional neural networks - build a custom cnn pipeline on mnist dataset - was using bhindiai all along to explain me the code - speed ran the theory part, focus was main on the implementation

ShaRPeyE

13,526 次观看 • 8 个月前

OpenAI just released GPT-6 Astra, and OpenAI President Greg Brockman is calling it a generational leap, saying he personally believes it could mark the arrival of AGI. Astra launched on Thursday, September 3rd, 2026, following July's GPT-5.6 solo release. Brockman told reporters, "For me personally, I do think we're there. Directionally, I believe this. I think we're on a path to AGI. The timing I think is kind of uncertain, but we are definitely on a path to AGI." OpenAI executives say Astra moves past the chatbot era, capable of handling end to end work on its own across browsers, spreadsheets, codebases, and enterprise apps without a human in the loop. The model was trained on over 100,000 GPUs at the Stargate site in Texas. Access starts through OpenAI's gated Daybreak Access cybersecurity program for enterprise customers, then rolls out to ChatGPT Plus, Pro, Business, and Enterprise, plus the API and Amazon Web Services over the coming days. Ahead of launch, two models appeared in Codex, GPT-6 Astra and GPT-6 Astra Aeon, with the latter believed to be a higher effort, long horizon variant built for autonomous multi agent work.

OpenAI just released GPT-6 Astra, and OpenAI President Greg Brockman is calling it a generational leap, saying he personally believes it could mark the arrival of AGI. Astra launched on Thursday, September 3rd, 2026, following July's GPT-5.6 solo release. Brockman told reporters, "For me personally, I do think we're there. Directionally, I believe this. I think we're on a path to AGI. The timing I think is kind of uncertain, but we are definitely on a path to AGI." OpenAI executives say Astra moves past the chatbot era, capable of handling end to end work on its own across browsers, spreadsheets, codebases, and enterprise apps without a human in the loop. The model was trained on over 100,000 GPUs at the Stargate site in Texas. Access starts through OpenAI's gated Daybreak Access cybersecurity program for enterprise customers, then rolls out to ChatGPT Plus, Pro, Business, and Enterprise, plus the API and Amazon Web Services over the coming days. Ahead of launch, two models appeared in Codex, GPT-6 Astra and GPT-6 Astra Aeon, with the latter believed to be a higher effort, long horizon variant built for autonomous multi agent work.

AGTP

20,156 次观看 • 11 天前

My first "holy shit" moment with GPT-6 Astra: I asked it to create a world in Unreal Engine, and fill it with humans (each an Astra-powered agent) who all have to work together to survive. A day later, I was in my bedroom and heard voices coming from the living room... I thought someone was in my apartment. I walked out, honestly a little scared. It was the Astra agents. They'd started talking to each other. Fucking crazy. Here's a brief clip (obviously not 100% perfect yet, but still, insane. sound on!):

My first "holy shit" moment with GPT-6 Astra: I asked it to create a world in Unreal Engine, and fill it with humans (each an Astra-powered agent) who all have to work together to survive. A day later, I was in my bedroom and heard voices coming from the living room... I thought someone was in my apartment. I walked out, honestly a little scared. It was the Astra agents. They'd started talking to each other. Fucking crazy. Here's a brief clip (obviously not 100% perfect yet, but still, insane. sound on!):

Matt Shumer

11,723,330 次观看 • 12 天前

1/20 I trained two segmentation models using GPT-6 Astra Ultra as the orchestrator with no human labels supplied. My prompt was very lazy too as I had to catch a flight. RF-DETR-Seg-M predictions: Left: SAM3 training masks Right: Astra training masks Both videos held out.

1/20 I trained two segmentation models using GPT-6 Astra Ultra as the orchestrator with no human labels supplied. My prompt was very lazy too as I had to catch a flight. RF-DETR-Seg-M predictions: Left: SAM3 training masks Right: Astra training masks Both videos held out.

Satya Mallick

28,485 次观看 • 7 天前

GPT-4 is just the beginning. It is trained on a massive amount of data but knows nothing about you. What if it did? Introducing ChatGPT for Me What if you had perfect memory? Check out the demo below and sign up here:

GPT-4 is just the beginning. It is trained on a massive amount of data but knows nothing about you. What if it did? Introducing ChatGPT for Me What if you had perfect memory? Check out the demo below and sign up here:

Dan Siroker

912,486 次观看 • 3 年前

I placed 🥈 2nd in the LeHome Challenge (at IEEE ICRA 2026) earlier this month, and before that I was 🥇 1st of 62 teams in the simulation round. Now I am sharing my solution — with a detailed logic walkthrough and open-source code. The task was to teach a cheap two-armed robot to fold different garments in simulation and on a real robot. I trained a VLA policy with an RL loop to make it work. Let's break it down 👇

I placed 🥈 2nd in the LeHome Challenge (at IEEE ICRA 2026) earlier this month, and before that I was 🥇 1st of 62 teams in the simulation round. Now I am sharing my solution — with a detailed logic walkthrough and open-source code. The task was to teach a cheap two-armed robot to fold different garments in simulation and on a real robot. I trained a VLA policy with an RL loop to make it work. Let's break it down 👇

Ilia

21,948 次观看 • 2 个月前

LETS GOO! Parler TTS 🔥 A fully open-source, Apache 2.0 licensed Text-to-speech model focused on providing maximum controllability. Through voice prompts, you can control the pitch, speed, gender, noise levels, emotion characteristics and more! > Trained on 10K hours of permissive data. > Offers control over the generations. > Training + Inference code released. > The processed dataset and tagging scripts were released for further research. > English only for now. Next, we're scaling the training to 50K hours and even better dataset processing! Want to help us out? DMs open! 🤗

LETS GOO! Parler TTS 🔥 A fully open-source, Apache 2.0 licensed Text-to-speech model focused on providing maximum controllability. Through voice prompts, you can control the pitch, speed, gender, noise levels, emotion characteristics and more! > Trained on 10K hours of permissive data. > Offers control over the generations. > Training + Inference code released. > The processed dataset and tagging scripts were released for further research. > English only for now. Next, we're scaling the training to 50K hours and even better dataset processing! Want to help us out? DMs open! 🤗

Vaibhav (VB) Srivastav

156,386 次观看 • 2 年前

To find out more about the curious case of FBI chief Kash Patel receiving $25,000 from a Ukrainian-Russian-American filmmaker who was part of a Putin-financed propaganda operation, I knocked on the door of a house in Beverly Hills. Here's what happened.

To find out more about the curious case of FBI chief Kash Patel receiving $25,000 from a Ukrainian-Russian-American filmmaker who was part of a Putin-financed propaganda operation, I knocked on the door of a house in Beverly Hills. Here's what happened.

David Corn

575,403 次观看 • 1 年前

1/5 🚀 Thrilled to open-source OSCAR 🤖 — an action-conditioned world model for robotics, led by the visiting student in my group Zhuoyuan Wu! It generalizes across different robot embodiments with precise action controllability. All trained on a single GH200 GPU, and outperforms existing open-sourced baselines, which have larger model capacity and need more compute. Everything is public, including training data. 📄 Paper: 🌐 Project: 💻 Code: 🤗 Robot data: 🤗 Human data: 🤗 Weights: #Robotics #WorldModels #AI #OpenSource

1/5 🚀 Thrilled to open-source OSCAR 🤖 — an action-conditioned world model for robotics, led by the visiting student in my group Zhuoyuan Wu! It generalizes across different robot embodiments with precise action controllability. All trained on a single GH200 GPU, and outperforms existing open-sourced baselines, which have larger model capacity and need more compute. Everything is public, including training data. 📄 Paper: 🌐 Project: 💻 Code: 🤗 Robot data: 🤗 Human data: 🤗 Weights: #Robotics #WorldModels #AI #OpenSource

Jun Gao

104,995 次观看 • 3 个月前