AI is more than LLMs
I feel like everybody knows this, but/and everybody (including me) is constantly forgetting it: AI is more than LLMs, and the really fun and weird stuff is tucked into that “more than”.
A few examples:
-
Are these new antibiotics produced with the help of AI tools really meaningful? I have no idea. Either way, don’t allow the LLM merchants any stolen valor —
these are totally different techniques. (This has consistently been the case for AI that’s actually useful in the lab, and each example ought perhaps to be read as a quiet strike against the premise that LLMs in their fruition will unlock new science.) -
Apple’s Depth Pro uses a single 2D image to generate a supersharp depth map: a plausible guess at the scene’s structure. The results are terrific, and a great starting point for 3D work. (TikTok’s Depth Anything is also very good, but I’ve found Depth Pro consistently more robust.)
-
Replicate’s Explore page is a vast AI bestiary, and it’s fascinating —
provocative — simply to browse and see all the things you could do with AI, right now, with a few bucks. For example, I found this AI model superfocused on refining manuscript images. There are a lot of cruddy image generators, sure … weird voice-cloners, too … and, for that reason, it’s useful to check in on this page every so often, just for the tour of present capabilities.