What if pretraining scientific foundation models didn’t require massive datasets at all?
We show that this is not only possible, but has a range of neat benefits: our "Tadpole" models learn from canonical PDE data that is generated on-the-fly https://t.co/yVWdTaYcqd
Excited to announce that the CAD rendering tool that I've been working on now supports STEP rendering. My goal was to make the fastest open source rendering tool available for agentic CAD, which seems to have been accomplished: look is 3x faster than OCCT for the model shown!
@emollick Curious to see how @reflection_ai does. They already pitch themselves as the frontier open model company for the US. They have the department of energy deal and probably going after top finance too
@melodyskim Yea, freaky to learn as I’ve been in tech but wasn’t the biggest X user. Once you find the algo, you realize how much of SV it moves. But I guess you can assume that from a public group chat
All of these token resets seems like a signal that “unlimited” is coming soon. Funny to be in the free calls after 6PM and weekends era knowing how we’ll forget it the same way