Pinned
Last year at @tryramp I laid out three predictions for how language models would evolve. I was trying to clarify which bets might actually be durable over time.
A lot of it is now starting to take shape.
Here’s an update. Thread 👇
- good separation of concerns for how ai for science will work in many domains imoa nice demonstration of Claude Science, but worth clarifying that the design is not "done by Claude" but by orchestrating tool calls of open-source, task-specific protein design models: PXDesign, RFdiffusion, Genie, BoltzGen, etc I think the direction of LLMs using
- morsel driven parallelism is the Harvard bar scene of good will hunting for database engineersIt's crazy how PlanetScale engineers will just casually drop a link to the most banger paper you've ever seen. A little database. A little query planning. A little NUMA-aware parallel scheduling. And as you might expect, a product of TUM scholars. 15721.courses.cs.cmu.edu/spring2016/pap…
- honestly text is just the wrong input for fine grained control of video generation, the sooner the video model providers adopt a more UX driven approach to deciding how to train the model the betterSeedance 2.5 is finally available in the US, but something feels off. I'm running comparisons to 2.0 and 2.5 is performing worse in most of my tests so far. Same start frames, references, and prompts. Multiple runs of each. 2.5's motion coherence, prompt adherence, visual
