applied healthbench + 230m users

Z / Runner 🏃🏻‍♂️

🏃🏻‍♂️ entry w/ shamir: OpenAi, HealthBench: (paper, code, blog), the core problem statement: crystallized medical knowledge is not useful, most AI fails on how it applied that knowledge in dynamic, unpredictable, often information-sparse clinical settings. (intelligence → decision → action are not connected in sota ai models). See analysis on Friday.

• 🏃🏻‍♂️ no diff on general agent & specific app; but need to know what are actionable (we can table this for today's sync).

• 🏃🏻‍♂️ want to know more, beyond the vague (my trading algo is awesome), for example, did aaron's 2025 ai trading algo net/net a gain or loss? I heard too many of the telltale stories of "ai trading bots" killing the market, but when pressing for the specifics, i get the general boiler plate response.

Stephen / Basketball 🏀

🏀 re: $20 vs nihilism: not all apps or features, but relatively and timing. hence, 230m users about (function) health, but only via chatgpt.

🏀 re: trading as agent architecture / showcase: ..

🏀 0%.