Ashish Rai

open source & open science AI // formerly @NYUniversity // perf engineering @VMware

[Ashish Rai @raish1sh]

here we go again :/

[Ashish Rai @raish1sh]

This

[ThePrimeagen @ThePrimeagen]

There is a certain side of running a business that these AI hypes are missing.

Always running at capacity leaves no room for really thinking about what is important

Sure the AI is "letting you do more than ever," but those same people stopped dreaming and are always executing.

Feels like a tragedy

[François Chollet @fchollet]

The best way to use AI is an interface to information that lets you deepen and improve your own knowledge and mental models. The worst way to use AI is as a crutch to outsource and forsake your own cognition

[Ashish Rai @raish1sh]

@__tinygrad__ @HDPbilly So finally AMD did send you some machines!

[tetsuo.mlir @tetsuo_cpp]

Oh, you're writing CUDA kernels? Everyone's on Triton now. Just kidding, we're all on Mojo. We're using cuTile. We're using ROCm. We have an in-house DSL compiler targeting the NVGPU MLIR dialect but wait, Tile IR just dropped so we're going to target that instead. Our PM is on TileLang. The team lead was on CuTe but now she's back to handwriting PTX. If you're not on Pallas, you're ngmi. Our intern is building on TT-Metalium for our Wormholes. Our CFO approved an order for some big chungus wafer-scale chips so now we're porting our kernels to CSL. Our CTO is working on a kernel-less graph compiler so we won't need to write kernels anymore. Our CEO thinks we're talking about the Linux kernel. We're building Claude for dogs.

[Ashish Rai @raish1sh]

@evilmathkid @mm_sasmitha 25-28% acc → 27% acc is not really an improvement in performance, this could very well be noise.

[Jared Roesch @roeschinc]

Thrilled to announce we're open-sourcing the CUDA Tile dialect and bytecode! https://t.co/wpy2BoybAk

What's included:
    • CUDA Tile MLIR dialect
    • Bytecode serialization/deserialization support
    • MLIR Python bindings for programmatic IR construction
    • Conformance test suite

For developers: You can now integrate CUDA Tile directly into your projects using MLIR and generate CUDA Tile dialect or bytecode natively!

Learn more about CUDA Tile:
    • NVIDIA Developer: https://t.co/vjf6KnrMMU
    • CUDA Tile Specification: https://t.co/QJiF8QVd2i

This project represents the collaborative effort of multiple teams across NVIDIA. A huge thanks to everyone who made this possible!

It has been a privilege to be involved.

[Ashish Rai @raish1sh]

"NVIDIA will continue to develop and distribute Slurm as open-source, vendor-neutral software" but it will run super-hyper-crazy fast on Nvidia chips 🐒

[Michael Black @Michael_J_Black]

Video diffusion models have strong implicit representations of 3D shape, material, and lighting, but controlling them with language is cumbersome, and control is critical for artists and animators.

GenLit connects these implicit representations with a continuous 5D control signal describing the direction and intensity of a point light source.

This enables single-image near-field relighting of an image using a video diffusion model. We use a ControlNet-like approach and show that, with a small amount of synthetic data, GenLit generalizes to complex real-world images.

Given a single image and the 5D lighting signal, GenLit creates a video of a moving light source that is inside the scene. It moves around and behind scene objects, producing effects such as shading, cast shadows, secularities, and interreflections with a realism that is hard to obtain with traditional inverse rendering methods.

GenLit shows that it is possible to get continuous control over implicit physical processes within a video model. I think this is just the beginning and promises to make such models much more practical for creators.

@sbharadwaj__ will present today at SIGGRAPH Asia Room: S423/S424, Level 4 @ 13:50 on 15 of Dec.

https://t.co/8fvLm8QUU7

[Ashish Rai @raish1sh]

Attended my first AI by hand talk for fun.
The choice of using Excel to demonstrate parallel computations is surprisingly intuitive. Great explanations by Prof Yeh in today's talk.

also TIL you can trace graphs between variables created using formulas in excel.

[Tom Yeh @ProfTomYeh]

Let's build a GPU by hand ✍️ together this Thursday 👉 Join my lecture https://t.co/ydmA0ypSkO

[Ashish Rai @raish1sh]

cool crowd, unfortunate photo timing lol

[Hannan @hannansyedshah]

cool to have learned about @Apple MLX, thanks to @a_yakovlev for the talk!

[Ashish Rai @raish1sh]

@ssahoo_ @GMartius @volokuleshov Congratulations! I have found your papers on diffusion language models to be some of the best ways to understand diffusion language modeling.

[Ashish Rai @raish1sh]

a great Monday afternoon read.

[Jascha Sohl-Dickstein @jaschasd]

Title: Advice for a young investigator in the first and last days of the Anthropocene

Abstract: Within just a few years, it is likely that we will create AI systems that outperform the best humans on all intellectual tasks. This will have implications for your research and career! I will give practical advice, and concrete criteria to consider, when choosing research projects, and making professional decisions, in these last few years before AGI.

This is my current go-to academic talk. It's mostly targeted at early career scientists. It gets diverse and strong reactions. Let's try it here. Posting slides with speaker notes...

--

The title is a play on a very opinionated and pragmatic book by the nobel prize winner ramon y cajal, who is one of the founders of modern neuroscience.

To get you in the right mindset, on the right we have a plot of GDP vs time.
That is you, standing precariously on the top of that curve.
You are thinking to yourself -- I live in a pretty normal world.
Some things are going to change, but the future is going to look mostly like a linear extrapolation of the present.

And the plot should suggest that this may not be the right perspective on the future.

This plot by the way looks surprisingly similar even if you plot it on a log scale. We didn't stabilize on our current rate of growth until around 1950.

[Goodfire @GoodfireAI]

Agents for experimental research != agents for software development.

This is a key lesson we've learned after several months refining agentic workflows!

More takeaways on effectively using experimenter agents + a key tool we're open-sourcing to enable them: 🧵

[Frédéric Legrand @FredericTheTall]

4 minutes of raw, 1x speed, piper robot arm being controlled by pi0.5, failures and successes

The model feels like a strong step up from pi0

[Ashish Rai @raish1sh]

Love the product https://t.co/Swh9rUOSCP is building!

I contributed to an initial version of supporting pi0 in the phosphobot platform: https://t.co/LgL4yynm5y

[Pierre-Louis Biojout (PLB) @plbiojout]

Try pi-05 from @physical_int today on your robot.

One click fine-tuning and inference from phosphobot.

No code and no GPU required.

-> docs .phospho .ai

[Decart @DecartAI]

When Zuck’s WiFi doesn’t work, Lucy Edit can express what he really thinks (yes, it can add text to clothing!)

[vik @vikhyatk]

Excited to release a preview of Moondream 3.

A 9B param, 2B active MoE vision language model that makes no compromises; offering state-of-the-art visual reasoning while still retaining an efficient and deployment-friendly form factor.

[Kimi.ai @Kimi_Moonshot]

Introducing checkpoint-engine: our open-source, lightweight middleware for efficient, in-place weight updates in LLM inference engines, especially effective for RL.

✅ Update a 1T model on thousands of GPUs in ~20s
✅ Supports both broadcast (sync) & P2P (dynamic) updates
✅ Optimized pipeline with overlapped communication and copy
✅ Lightweight & flexible for large-scale deployment

Check out our work on GitHub: https://t.co/L4c1oUAB6p

[Ashish Rai @raish1sh]

@OBalcells I think this would be worthwhile.

https://t.co/XGZrJqFtYx

[Ashish Rai @raish1sh]

a good exercise would be to use the probe classifier scores to guide reasoning traces in small language models and compare it to KD from reasoning model traces.