Omar Khattab (@lateinteraction) / X

Omar Khattab

14.1K posts

Omar Khattab

@lateinteraction

asst professor @MIT CSAIL @nlp_mit. ColBERT.ai, DSPy.ai (@DSPyOSS), GEPA, RLMs, Pedagogical RL

Cambridge, MA

Joined December 2022

Omar Khattab
@lateinteraction
7h
OK but that “without symbols” bothers me 😅. We definitionally created RLMs as (1) “symbolic recursion” over (2) symbolic references to the prompt. RLMs excel at compositional generalization *because* they’re neurosymbolic. Good luck getting a vanilla Transformer to match that.
wh
@nrehiew_
14h
95% AHAHHAA Super cool release but can we just all ignore ARC-AGI forever now? “ARC-AGI 4 will definitely be unsolvable without symbols! This time for sure!”
Omar Khattab
@lateinteraction
13h
haven’t been doing much of the last one on that list for a few months wow
Omar Khattab
@lateinteraction
Feb 20
i'm a simple guy, i log in, retweet the latest colbert model, quote the latest gepa update, amplify the latest RLM usecase, like the latest product blog on using dspy, write a quick pun or two, and log out
Omar Khattab
@lateinteraction
18h
user-facing RLMs !
Prime Intellect
@PrimeIntellect
19h
Introducing Prime Agent: A self-improving RLM harness for coding and long-running autonomous tasks. Designed to be both token-efficient and expressive through programmatic tool calling, context as a variable, multi-agent messaging, and a self-modifiable harness state.
00:00
Omar Khattab
@lateinteraction
18h
RLMs crushing it.
Prime Intellect
@PrimeIntellect
19h
Replying to @PrimeIntellect
Prime Agent is a general-purpose coding harness On ARC-AGI-3, it scores 95.5%, surpassing the human-expert baseline, but the gain is not benchmark-specific. We see major improvements across models when compared to their proprietary harnesses:
Omar Khattab
@lateinteraction
Aug 4
man, we’re so extremely lucky to have @isaacbmiller1 as lead maintainer! also wow it took me longer than i’d like to admit to realize that the list below shows only the *first-time* contributors who contributed to 3.3.0! many familiar names. thank you folks for contributing.
Isaac Miller 🧩
@isaacbmiller1
Aug 3
DSPy 3.3.0 is out! 3.3.0 expands what DSPy can optimize and how it connects to LMs: - dspy.Flex is a Module that allows GEPA to optimize the code, as well as the prompt, of a program. - dspy.ReActV2 adds native and parallel tool calling over dspy.ReAct - A new typed,