Uri Katz

NLP PhD student @biunlp. Working on IR and Search, QA Reasoning. Intern at @allen_ai
🍄‍🟫🌵🍄🍄‍🟫🌵🍄🍄‍🟫🌵🍄

[Uri Katz @UrikaUri]

🐳Introducing Knowledge Navigator🐳: A new way to explore scientific literature! Our paper shows how LLMs can transform information overload into structured, navigable knowledge.

paper: https://t.co/VKQAmuq7il
page (with demo): https://t.co/H0mOf1oFwf

🧵/9

[Matan Avitan @matan_avitan_]

🚨 We present 𝐌𝐀𝐍𝐂𝐄: 𝐌𝐀𝐍𝘪𝘧𝘰𝘭𝘥 𝘢𝘸𝘢𝘳𝘦 𝐂𝘰𝘯𝘤𝘦𝘱𝘵 𝐄𝘳𝘢𝘴𝘶𝘳𝘦 (with @yoavgo and @yanaiela).

MANCE is a new state-of-the-art method for concept erasure.

[Uri Katz @UrikaUri]

@hxiao @lrzneedresearch Really one of the best events in emnlps, they rejected it for emnlp 2026?

[Uri Katz @UrikaUri]

@yoavgo Maybe authors rejection for a full year will do the work. Id like to see senior researchers risk their lab. Also maybe it will encourage them to read student's submissions

[Uri Katz @UrikaUri]

@boknilev @itayevron yes i get the same feeling when reviewing but also when trying "vibe research" by myself. Like claude is chasing small experimental "rewards", overcooking meaningless findings.. need to find a way to tame it

[Uri Katz @UrikaUri]

@CENTCOMHebrew למה לכולם אצלכם יש שפם?

[Mosh Levy @mosh_levy]

New paper!
People treat reasoning trajectories as text, but what if we can do better than that?
We show that we can, by training Behavior Forecasters (BFs) that get a reasoning trajectory as input and make more accurate forecasts than frontier models at a fraction of the cost. 🧵

[Uri Katz @UrikaUri]

@Eyal__Weiss למה מלכתחילה סטודנטים צריכים ללמוד מאמרים שלך?

[Uri Katz @UrikaUri]

@cohenrap @CiItay @boazbe בנזונה בהיי��ק

[Uri Katz @UrikaUri]

@shareyourgold At least I have chicken

[Ai2 @allen_ai]

We analyzed 250K+ queries & 430K+ clickstream interactions from Asta, our AI-powered research assistant—and today we're releasing the full dataset. How do researchers actually use AI science tools? Here's what we found. 🧵

[Elron Bandel @ElronBandel]

🧵 Should AI agents be specialized or general purpose?

We think this question defines the Linux era for AI.
@AsafYehudai @evijit @alex_lacoste_ @gneubig @mmitchell_ai @michalshmu @LChoshen

[intern @intern]

if your friends aren���t talking about:

- claude code
- creatine
- openclaw
- looksmaxxing
- ai agents
- taste
- prediction markets
- mac minis

you should be grateful & cherish those people

[Uri Katz @UrikaUri]

@dan_derazne @yuvalbenj @ZuckermanRoy התנצלתי מראש.

[Uri Katz @UrikaUri]

@yuvalbenj @dan_derazne @ZuckermanRoy סליחה כן?

[Ai2 @allen_ai]

Introducing Theorizer: Turning thousands of papers into scientific laws 📚➡️📜

Most automated discovery systems focus on experimentation. Theorizer tackles the other half of science: theory building—compressing scattered findings into structured, testable claims. 🧵

[Uri Katz @UrikaUri]

@UrbanElegance22 @niron תפרט? אשמח לשמוע יותר

[Uri Katz @UrikaUri]

Is that a ragebait??

[OpenAI @OpenAI]

We view chain-of-thought monitoring as complementary to mechanistic interpretability, not as a replacement for it.

Because we believe that chain-of-thought monitoring is incredibly useful as a window into a model’s brain and could be a loadbearing layer in a scalable control scheme, we plan to expand upon these evaluations and use them to inform future modeling and data decisions.

[Uri Katz @UrikaUri]

@FinancialEduX @GrimbergYuval @urieli17 תושבי גוש דן לא בנויים לחומרה.

[Mosh Levy @mosh_levy]

New paper:
We are often told that reasoning tokens aren't faithful explanations. But to have a useful metaphor for their operation we need a characterization of what they are, not what they are not.
To that end, we suggest "State over Tokens" (SoT) 👇🧵