Keep on to blur preview images; turn off to show them clearly
Google 玩明白了
行道途中。非求速成，惟求通达。 2023 年扎进AI ，打通Know-How，不少赚钱项目，踩过坑，也见过光。 围城里待得够久了，出来聊聊世界，聊聊技术、聊聊赚钱。
RT @bradenjhancock: Today we announced Slingshots // ONE, the first batch of Laude Slingshots grant recipients. And dang — it’s an impressi…
Asst professor @MIT EECS & CSAIL (@nlp_mit). Author of https://t.co/VgyLxl0oa1 and https://t.co/ZZaSzaRaZ7 (@DSPyOSS). Prev: CS PhD @StanfordNLP. Research @Databricks.
RT @getpy: DSPyWeekly Issue #10 is live! This issue covers: - Articles: Bay Area DSPy Meetup, new papers (E-CARE, Agent-Omni), using ReA…
Asst professor @MIT EECS & CSAIL (@nlp_mit). Author of https://t.co/VgyLxl0oa1 and https://t.co/ZZaSzaRaZ7 (@DSPyOSS). Prev: CS PhD @StanfordNLP. Research @Databricks.
The new design is more cluttered, but I think it is more useful. Which do you think is better?
Productivity meets minimalism. Building, designing, and sharing my solutions.
> the curvature for memorized training points is much sharper than non-memorized, meaning ordering weight components from high to low curvature can reveal a distinction without explicit labels I've been talking of «spiky singularities» forever, they treat this formally
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
Enough geopolitics, this is an extremely interesting advance in mech interpretability. Goodfire uses loss curvature analysis to decompose memorized and generalized structures on the level of weight-space, and suppress memorization via weight-editing.
> the curvature for memorized training points is much sharper than non-memorized, meaning ordering weight components from high to low curvature can reveal a distinction without explicit labels I've been talking of «spiky singularities» forever, they treat this formally