Finally Harvard lives up to its motto: Veritas view.hu.harvard.edu?qs=351b45ebf...

Simulated AI hospital where “doctor” agents work w/ simulated “patients” & improve: “After treating around ten thousand patients, the evolved doctor agent achieves a state-of-the-art accuracy of 93.06% on a subset of the MedQA dataset that covers major respiratory diseases” arxiv.org/abs/2405.02957

Reasoning AI models require training on human reasoning. One of the real gaps in pushing forward these models is going to be the old problem of how to figure out how to get experts to explain what they do.

AI keeps bumping up against our limited knowledge of how expertise works

“GPT-4o, o1, o1-preview & o1-mini all demonstrate strong persuasive argumentation abilities, within the top ~80-90% percentile of humans (i.e., the probability of any given response from one of these models being considered more persuasive than human is ~80-90%)”

-o1 system card

Veo 2 prompt: "a distant shot zooms in to reveal a knight wearing a golden helmet, he begins to charge on his zebra, lowering his lance, charging towards a clockwork octopus" (this is one of the initial 4 videos it made)

I wrote about what was important in a crazy month of AI advances:
1) Intelligence, of a sort, everywhere
2) The first very smart models that can push us to push the boundaries of knowledge
3) AI got eyes to go with its ears
4) Leaps in video generation
www.oneusefulthing.org/p/what-just-...