Hen Davidov

Rhodes Scholar, StatML CDT @UniOfOxford |
@Forbes 30U30 2025 |
Ex- @AmazonScience , @TechnionLive

[Hen Davidov @hendav136]

Recently accepted to ICML 2026: Knowing when to quit. We refuse to complete failing LLM generations as they unfold, token by token, instead of judging the prompt up front or scanning the final output.
The method utilizes a 2-layer probe to estimate expected correctness.

[Hen Davidov @hendav136]

@ReddyAarya36823 @noclearnogga Lol dope

[Hen Davidov @hendav136]

Hen, first year PhD, pretraining an LLM with academia compute for the first time.
Wish me luck.

[Hen Davidov @hendav136]

@redtachyon The Fable nerf came right after the first tweet

[Hen Davidov @hendav136]

Someone at anthropic saw my tweet I guess? Can't explain the timing any other way

[Hen Davidov @hendav136]

Hen, first year PhD, pretraining an LLM with academia compute for the first time.
Wish me luck.

[Hen Davidov @hendav136]

Someone at anthropic saw my tweet I guess? Can't explain the timing any other way

[Hen Davidov @hendav136]

@ThakurAryancha4 Best of luck with that!

[Hen Davidov @hendav136]

@ThakurAryancha4 Starting with a tiny gpt3-esque 100m param, and building up on the positive signals

[Hen Davidov @hendav136]

@noclearnogga Let's start by changing your profile pic from a dude with a swastica on his forehead?

[Hen Davidov @hendav136]

@mesken_stefan Thank you!!

[Hen Davidov @hendav136]

@ruth_rallph Thank you!

[Hen Davidov @hendav136]

@JerrryKun I have a transformers arch bet in mind, hoping to make a dent in this possibly saturated line of work

[Hen Davidov @hendav136]

@gradascetic Less than you'd expect. Compute is kinda siloed off in different labs. I mainly use my lab's gpus

[Hen Davidov @hendav136]

@ZacharyBamberg1 Thank you! Right now I'm just working off of a llama style recipe, iterating on my arch bet

[Hen Davidov @hendav136]

@yoav_gelberg I'm not in the kindergarten bro lets be real

[Hen Davidov @hendav136]

@Besteuler Thank you!

[Hen Davidov @hendav136]

God I love science sometimes

[Eric Topol @EricTopol]

There hasn't previously been a treatment vs pancreatic cancer this successful. Striking improved (a > doubling) survival results @NEJM and @ASCO today with daraxonrasib, which also became available via an FDA approved early access program and began shipping to physicians this week @RevMedicines
https://t.co/e04jqJMPw0

[Hen Davidov @hendav136]

@klachkin @Hadionysy וגם לא נעים להגיד היה נורא לנשים

[Hen Davidov @hendav136]

@eygopt לכל הפעולות היה פירוק ספקטרלי משותף, וכנראה היה מתקיים P=NP, אז כן מסכים מאוד

[Elizabeth Barnes @BethMayBarnes]

Sometimes people outside the field say things like “The AI situation can’t be that bad, there must be experts who are on top of it”. As “an expert”, I would like to be clear that we are *not* on top of it. Some key aspects of the situation IMO: