Kunal retweeted
Intelligence is about learning - from few examples and from interaction with the world. Current models are not intelligent. They are very capable and useful, but not intelligent in the way humans are because they need so many examples to reach human level.
LLMs are not bitter lesson pilled because they still rely on human data, and lots of it. An intelligent agent- a truly intelligent agent- can just learn from the world by continual exploration.
I think this is what he means.
Kunal retweeted
we’re doing cool things. join us in delivering science fiction
tesla.com/careers/search/job…
Replying to @teslayoda
We’re making interesting progress here.
Digital Optimus can play about halfway through the Diablo campaign so far just by looking at the screen like a human.
Skill at Counter-Strike and other fast games is good too.
League is also in training.
Goal is to generalize across all games.
We just released a new challenge today ❗
GPT-2 (120M) Transformer Block
Compose multiple kernels into a full transformer block. The first of many upcoming challenges focused on real-world inference optimization.
Write your solution in CUDA, Triton, PyTorch, JAX, Mojo, or CuTe DSL and benchmark it on state-of-the-art GPUs like the H100, H200, and B200, and more.
Kunal retweeted
Replying to @mysticwillz
API gets 10x traffic overnight with no page? I'm checking revenue! 😆
Kunal retweeted
Oh so that's who's #1 on our leaderboards 😳
Kunal retweeted
Replying to @sadernoheart @LeetGPU
@LeetGPU is the best place to build muscle memory on kernel writing, i dont say that lightly, your not a cuda monkey if your not leetgpu maxxing!!!
Kunal retweeted
Neat website that has a bunch of CUDA programming problems:
leetgpu.com/
Kunal retweeted
I think @xai should be recruiting from LeetGPU's userbase
Kunal retweeted
We just hit 100K code submissions across Playground & Challenges. The GPU programming community is growing as fast as ever! 🚀
Next stop: 1M in the next 3 months.
Kunal retweeted
Many people assume that LRM reasoning breaks down past a certain "complexity" or "number of steps" threshold. This is incorrect. It breaks down past an unfamiliarity threshold. And that threshold is very low.
There is no limit to the complexity of tasks you can solve with these models, no limit to the number of steps in the reasoning chains they can master -- as long as they have been covered during training / tuning. However, show them something unfamiliar, even very simple and requiring just a handful of reasoning steps (e.g. an ARC 2 task), and they will fail.
The reason you see a steps/complexity threshold when testing on problems like Towers of Hanoi is because these are familiar problems. To see a breakdown you need to turn such a familiar task into a novel one, and one way to achieve this is to scale up the problem variables. The complexity increase is just a roundabout way to generate novelty.
It's not about simple vs complex. It's familiar vs novel. Always has been.
Kunal retweeted
Global Leaderboards are here! 🏆🌎
You can now see the top 100 performers across languages and try to climb to the top!
Kunal retweeted
We just crossed 2000 users!! Thank you all so much for your support. We are going to be shipping some amazing & exciting things in the coming weeks 🚀🎉





















