Loading video...

Video Failed to Load

Go Home

My first PhD paper!🎉We learn *diffusion* models for code generation that learn to directly *edit* syntax trees of programs. The result is a system that can incrementally write code, see the execution output, and debug it. 🧵1/n

shreyaskapur's profile picture

Shreyas Kapur

2,576 subscribers

742,386 views • 2 years ago •via X (Twitter)

Science & TechnologyEducationNews & Politics
LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Excited to introduce Diffusion Augmented Agents (DAAGs)✨. We give an agent control of a diffusion model, so it can create its own *synthetic experience*.🪄 The result is a lifelong agent that can learn new reward detectors and policies, much more efficiently. Here's how. 👇
0:18

Excited to introduce Diffusion Augmented Agents (DAAGs)✨. We give an agent control of a diffusion model, so it can create its own *synthetic experience*.🪄 The result is a lifelong agent that can learn new reward detectors and policies, much more efficiently. Here's how. 👇

Norman Di Palo

12,874 views • 1 year ago

Today we're launching new ways to learn in Claude Code and the Claude app. First up: Claude Code now lets you customize communication styles with /output-style
0:19

Today we're launching new ways to learn in Claude Code and the Claude app. First up: Claude Code now lets you customize communication styles with /output-style

Claude

705,097 views • 9 months ago

You can edit your source code directly from Chrome Sources tab and it will synchronize with the filesystem! That HMR is dope! Chrome Angular
0:11

You can edit your source code directly from Chrome Sources tab and it will synchronize with the filesystem! That HMR is dope! Chrome Angular

Enea Jahollari 🅰

11,620 views • 8 months ago

Can we use motion to prompt diffusion models? Our #ICLR2024 paper does just that. We propose Motion Guidance, a technique that allows users to edit an image by specifying “where things should move.”
0:15

Can we use motion to prompt diffusion models? Our #ICLR2024 paper does just that. We propose Motion Guidance, a technique that allows users to edit an image by specifying “where things should move.”

Daniel Geng

16,304 views • 2 years ago

🚀 Introducing GenLit – Reformulating Single-Image Relighting as Video Generation! We leverage video diffusion models to perform realistic near-field relighting from just a single image—No explicit 3D reconstruction or ray tracing required! No intermediate graphics buffers, directly in the pixel space! 📄 Dive into the paper: 🎥 Project page & demos: 🛠 Code coming soon! #GenerativeAI #ComputerVision #Relighting #DiffusionModels #Graphics 🧵 1/5
0:19

🚀 Introducing GenLit – Reformulating Single-Image Relighting as Video Generation! We leverage video diffusion models to perform realistic near-field relighting from just a single image—No explicit 3D reconstruction or ray tracing required! No intermediate graphics buffers, directly in the pixel space! 📄 Dive into the paper: 🎥 Project page & demos: 🛠 Code coming soon! #GenerativeAI #ComputerVision #Relighting #DiffusionModels #Graphics 🧵 1/5

Haiwen (Haven) Feng

22,392 views • 1 year ago

We're also excited for Canvas to become a more personalized tutor that you can interact in much richer ways. You can ask ChatGPT to explain some mathematical concept, write code to plot that will help you learn it in a more visual way.
0:16

We're also excited for Canvas to become a more personalized tutor that you can interact in much richer ways. You can ask ChatGPT to explain some mathematical concept, write code to plot that will help you learn it in a more visual way.

Karina Nguyen

59,801 views • 1 year ago

1/ Happy to share VADER: Video Diffusion Alignment via Reward Gradients. We adapt foundational video diffusion models using pre-trained reward models to generate high-quality, aligned videos for various end-applications. Below we generated a short movie using VADER 😀, we used ChatGPT to write a script and an off-the-shelf AI music generator to generate the sound. Our code & weights are open-sourced:
0:16

1/ Happy to share VADER: Video Diffusion Alignment via Reward Gradients. We adapt foundational video diffusion models using pre-trained reward models to generate high-quality, aligned videos for various end-applications. Below we generated a short movie using VADER 😀, we used ChatGPT to write a script and an off-the-shelf AI music generator to generate the sound. Our code & weights are open-sourced:

Mihir Prabhudesai

13,330 views • 1 year ago

We have released the code and weights for our #CVPR2023 paper "Avatars Grow Legs: Generating Smooth Human Motion from Sparse Tracking Inputs with Diffusion Model"! code: abs: project: The demo is below:
0:14

We have released the code and weights for our #CVPR2023 paper "Avatars Grow Legs: Generating Smooth Human Motion from Sparse Tracking Inputs with Diffusion Model"! code: abs: project: The demo is below:

Artsiom Sanakoyeu

35,710 views • 3 years ago

My Three.js + TSL shader course — Fragments — is open for registration for the next 2 weeks! It's the perfect way to learn the techniques that power some of the most incredible shaders out there, and how to build them in code.
0:16

My Three.js + TSL shader course — Fragments — is open for registration for the next 2 weeks! It's the perfect way to learn the techniques that power some of the most incredible shaders out there, and how to build them in code.

phobon

13,671 views • 3 months ago

Meta just announced that Code Llama was now free for both research and commercial. This might the strongest competitor to ChatGPT: ▸ Can generate, explain, and debug your code ▸ Handles input 100,000 tokens ▸ Free for research + commercial use ▸ Outperforms most open models ▸ Comes in 7B, 13B, and 34B ▸ Supports Python, C++, Java, PHP, Typescript (Javascript), C#, and Bash Available in: ▸ Foundation base models (Code Llama) ▸ Python specializations (Code Llama - Python), ▸ Instruction-following models (Code Llama - Instruct)
0:11

Meta just announced that Code Llama was now free for both research and commercial. This might the strongest competitor to ChatGPT: ▸ Can generate, explain, and debug your code ▸ Handles input 100,000 tokens ▸ Free for research + commercial use ▸ Outperforms most open models ▸ Comes in 7B, 13B, and 34B ▸ Supports Python, C++, Java, PHP, Typescript (Javascript), C#, and Bash Available in: ▸ Foundation base models (Code Llama) ▸ Python specializations (Code Llama - Python), ▸ Instruction-following models (Code Llama - Instruct)

Lior Alexander

196,384 views • 2 years ago

Diffusion models are great, but we can squeeze out so much more from them. The only problem is that it usually requires extra training or manual representation editing. In our new paper, we show that with the current capabilities of LLMs, it is much simpler than we thought!
0:17

Diffusion models are great, but we can squeeze out so much more from them. The only problem is that it usually requires extra training or manual representation editing. In our new paper, we show that with the current capabilities of LLMs, it is much simpler than we thought!

Yossi Gandelsman

34,510 views • 2 months ago

AI has huge potential to change how we work and live. That was clear at Montreal’s ALL IN Conference. But we also need to make sure our AI ecosystem can develop responsibly – and that’s what our new code of conduct does. Learn more about the code here:
0:19

AI has huge potential to change how we work and live. That was clear at Montreal’s ALL IN Conference. But we also need to make sure our AI ecosystem can develop responsibly – and that’s what our new code of conduct does. Learn more about the code here:

Justin Trudeau

347,543 views • 2 years ago

Robotics has a dirty secret. The biggest bottleneck isn't hardware. It isn't compute. It's data. And right now, only a handful of corporations can afford to generate it... $ROBA No Code Studio is about to change that. 🧵(1/8)
0:15

Robotics has a dirty secret. The biggest bottleneck isn't hardware. It isn't compute. It's data. And right now, only a handful of corporations can afford to generate it... $ROBA No Code Studio is about to change that. 🧵(1/8)

Roba Labs

23,621 views • 1 month ago

But can Sora write code? "have someone explain how to write python code on a black board that outputs "hello world""
0:10

But can Sora write code? "have someone explain how to write python code on a black board that outputs "hello world""

Matthew Berman

377,625 views • 8 months ago

🤖 Try Autopilot (Preview) in VS Code! With Autopilot enabled, the agent stays in control of the workflow. It can run tools, retry on errors, and continue working until the task is complete. Learn more:
0:11

🤖 Try Autopilot (Preview) in VS Code! With Autopilot enabled, the agent stays in control of the workflow. It can run tools, retry on errors, and continue working until the task is complete. Learn more:

Visual Studio Code

109,440 views • 2 months ago

we open sourced the code to transform human videos into robot trajectories, so you can train robots with your hands 👐🏻 we used it in our recent paper R+X: Retrieval and Execution from Everyday Human Videos (ICRA 2025 🇺🇸) link and details in thread 🧵
0:11

we open sourced the code to transform human videos into robot trajectories, so you can train robots with your hands 👐🏻 we used it in our recent paper R+X: Retrieval and Execution from Everyday Human Videos (ICRA 2025 🇺🇸) link and details in thread 🧵

Norman Di Palo

10,578 views • 1 year ago

🌐 Agentic Browser Tools (Experimental) in Visual Studio Code! Agents can now open pages, read content, click elements, and verify changes directly in the integrated browser while building your web app. Enable ⚙️ workbench.browser.enableChatTools to try it out. Learn mode:
0:19

🌐 Agentic Browser Tools (Experimental) in Visual Studio Code! Agents can now open pages, read content, click elements, and verify changes directly in the integrated browser while building your web app. Enable ⚙️ workbench.browser.enableChatTools to try it out. Learn mode:

Visual Studio Code

211,005 views • 2 months ago

Don't trust, Verify ! You can now dig into the core building blocks of Keet to verify the code and cryptography that is being used ! Visit & to view the first wave of our open-source code ! More updates and code to follow soon!🍐
0:15

Don't trust, Verify ! You can now dig into the core building blocks of Keet to verify the code and cryptography that is being used ! Visit & to view the first wave of our open-source code ! More updates and code to follow soon!🍐

Holepunch🕳🥊

140,249 views • 3 years ago

How well can Qwen3.5 models debug code? I built BugFind-15 — 15 buggy snippets across Python, JS, Rust, and Go. Docker sandbox compiles and validates every fix. Two trap scenarios where the code is correct and the model must resist "fixing" it. Tested every Qwen3.5 size from 0.8B to 397B, plus Jackrong's popular distilled model (V2). The 0.8B scored 5%. The 2B scored 10%. At 4B, debugging ability jumps to 69%. The hardest scenario: BF-03, a Rust trap. The code compiles fine — format! borrows, it doesn't move. Not a single model figured this out. From 0.8B to 397B, every one of them "fixed" a bug that doesn't exist. Category C (subtle bugs — mutable defaults, integer overflow, slice aliasing) was 100% across every model 4B and above. Category D (red herring resistance) told the real story — can it resist fixing code that isn't broken? No model scored above 90%. Small models can't debug. Mid-size models fix obvious bugs but fall for traps. Large models fix the hard bugs but still invent problems that don't exist.
0:11

How well can Qwen3.5 models debug code? I built BugFind-15 — 15 buggy snippets across Python, JS, Rust, and Go. Docker sandbox compiles and validates every fix. Two trap scenarios where the code is correct and the model must resist "fixing" it. Tested every Qwen3.5 size from 0.8B to 397B, plus Jackrong's popular distilled model (V2). The 0.8B scored 5%. The 2B scored 10%. At 4B, debugging ability jumps to 69%. The hardest scenario: BF-03, a Rust trap. The code compiles fine — format! borrows, it doesn't move. Not a single model figured this out. From 0.8B to 397B, every one of them "fixed" a bug that doesn't exist. Category C (subtle bugs — mutable defaults, integer overflow, slice aliasing) was 100% across every model 4B and above. Category D (red herring resistance) told the real story — can it resist fixing code that isn't broken? No model scored above 90%. Small models can't debug. Mid-size models fix obvious bugs but fall for traps. Large models fix the hard bugs but still invent problems that don't exist.

stevibe

35,006 views • 2 months ago