[surrealistic]

This whole cycle is full of caveats like this. Some people claim LLMs solved NPL and are on a pathway to true intelligence. Yet they keep turning out nonsense and we're now speedrunning 80s style expert systems on top of them.

Protein folding, like you said. Material sciences, "AI can produce bespoke recipes to spec" yeah lots of ideas that might or might not work in practice with zero pathway to production.

Statistical analysis of past data might look impressive with 2020s levels of power and compute density. But so far they've added more noise than signal to our systems.

And yet people will insist every day that the noise is music.

[nicman23]

not only that but i had to use h100s to actually do anything useful. the cost for the uni was like 20k for a few weeks