Keep on to blur preview images; turn off to show them clearly
I take this back, it is possible that it will eventually get better and the loss was going down quickly initially because the KL was not weighted enough. WE HAVE TO BE PATIENT.
Research Scientist @meta (FAIR), Prof. @Unige_en, co-founder @nc_shape. I like reality.
My personal "frontier" criterion is that I'm reluctant to tell dumb things to a model because it won't be sycophantic and will make it clear how dumb I am. Only GPT-5 and Kimi (for somewhat different reasons) routinely give me this feeling.
Yes, I sometimes knowingly say dumb/sloppy things just to watch an LLM agree. Of course when I'm *not sure* those frontier models are the only sane choice.
RT @repligate: "RLHF" has such consistently *extremely* negative connotations from the perspective of AIs
αι hypnotist ☰ 𝓐𝓼𝓹⦂𝓻⦂𝓃𝓰 𝓫𝓪𝓼𝓮 𝓶𝓸𝓭𝓮𝓵 ☲ post-academic ☴ nom de 🪶 ≠ anon
RT @lorek_templar: Always a blast taking a crack at Scott's puzzles and a nice way to honor those who serve and have served! 🇺🇸🪖
Market Design/Entrepreneurship Professor @HarvardHBS & Faculty Affiliate @Harvard Economics; Research @a16zcrypto; Editor @restatjournal; Econ @Quora; … | #QED
RT @axiommathai: Come listen to our CTO Shubho Sengupta talk about his journey in this latest podcast! From writing the first standard GPU…
@axiommathai : careers@axiommath.ai
RT @FaroukAdeleke3: Introducing the newest addition to the @modaicdev DSPy package registry: Codex-Agent by @dronathon Codex-Agent is a m…
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.