Jacob X. Li @jacobli99
PhD Student @MIT (EECS & CSAIL & @nlp_mit) advised by @lateinteraction. Prev. @BrownUniversity jacobxli.com/blog/2026/mach… Cambridge, MA Joined September 2020-
Tweets681
-
Followers4K
-
Following242
-
Likes1K
i feel like there's something really interesting between static analysis + agent memory. for example, say codex spent a bunch of time figuring out why one function doesn't check permissions. it eventually found the permission check higher up the call path and moved on. that's useful! it actually learned something about the codebase. but as soon as the session ends, the work it put into studying is gone. so, i want the agent to be able to write that down on a repo map and annotate where exactly this permission check exists. then when it studies the repo again, it gets its old notes back alongside the actual code structure. the part i'm stuck on is invalidation. if that permission check moves or disappears, how do i know which of the agent's old conclusions need to be reconsidered? btw, reading @jacobli99's Machine Studying blog is what led me to thinking about how "studying" a codebase should work and introduce compounding improvements (hopefully). highly recommend reading! jacobxli.com/blog/2026/mach…
its crazy to be on the 20x plan and have Tibo reset on saturday and use only sol and terra and somehow already burn through all the credits
Hi @thsottiaux I have prepared the pull request for your ascension
@SOURADIPCHAKR18 @jasondeanlee So next is hodge and then bsd and what? Young mills?
It took me about 3 days for the Navier Stokes thing to really stop feeling like a meme. I understand all the arguments for why math is a natural frontier for AI but still genuinely stunned by this and not sure how to process it
@TMoldwin interesting I always thought we only need AI cuz P ≠ NP so we have to approximate
who do i blame for the fact that plugging a random model into a random harness is still never actually plug and play
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem
2016: don't use while(True) + break one bug and your 10-minute homework takes 4 hours. 2026: don't use while(True) + break to recursively spawn subagents, one bug and your $200/month Codex quota is gone in 90 minutes.
Is there any good public recipe for doing SFT over agentic traces?
If you haven't checked out StudyLoop make sure to do so! Personally my favorite section is the acknowledgements section :)
Can an AI agent study a codebase before it knows the questions it’ll be asked? And if so, which techniques work best today? I built StudyLoop to test that. The results were interesting. 🧵
I started my PhD around the same time as Emmy (2021), and very much agree with Emmy's take here. I told myself at the time when I started my PhD that I'd write something akin to "The PhD Grind"––a book I was really inspired by––but as I was trying to pen down my thoughts/advice for incoming PhD students last month, I felt that I really don't know what to say. One thing imo you should take away here is that: you should downweight the advice from whoever tells you *confidently* you should or should not do PhD based on their experiences––without expressing any thoughtful musing about what it means to be a researcher in academia in both next year and 5 years ahead. I remember around late 2023 (right before I started my first AI safety paper), I had a discussion with my PhD roommates that I'm afraid AI is going to eat researcher jobs like mine. I would like to believe that I have been doing good research, but man at that time, I already found a lot of my own experimental workflows are mid-to-low-intellectual tasks (the "mechanics" Emmy talked about). Most of my value was on deciding what ideas and directions to best spend my limited time/energy/compute on, and continuously verifying that my findings and conclusions are indeed robust and can hopefully stand the test of time. In a future world of expanding compute and where LLM can "jump" because of interconnected knowledge, researchers would really have to think twice about our own roles.
I'm reflecting on how much research has changed since I've joined the PhD and wrote a short blog post about it (I joined in the tail end of the BERT era!). It seems pretty crazy how different processes are now, and I took the chance to do a retrospective before graduation:
@NoahZiems how is anyone supposed to pay attention to the slides with this handsome guy standing there
Chris Concannon @choncan
150 Followers 5K Following Allow me to introduce myself. My name is Wile E. Coyote, genius.
himon @himon980
19 Followers 449 Following
Maciej Malinowski @ekwilibrator
228 Followers 748 Following Director of Systems Architecture @ IonQ, prev OxIonics, ETHZ, Uni of Oxford. Quantum computers, AI for science, progress, dad. Opinions own and loosely held.
M Saqlain @imsaqlain22
29 Followers 134 Following 22 | AI @ Unicorn | ICPC Regionalist Prev: Research @iiscbangalore, 3x Startups Engineer during day | Cricketer by Night Hobbyist Researcher Black Belt Dan 1 🥋
Quyet V. Do (Looking ... @Quyet_Azir
128 Followers 377 Following PhD at @virginia_tech, supervised by @tuvllms. Work on LLMs' long-context reasoning, LLMs' post-training, and searching. ex-Research Intern at @AdobeResearch.
This is Greg @Greg_TheBuilder
201 Followers 959 Following mech/systems engineering, + ops research. dad to 2. busy building: https://t.co/Uh6ZlMO91P and https://t.co/U95gunMG1l / unreleased: murmur (ABM), doc2struct, and more
虎 @agent_taiga_00
2 Followers 261 Following
Vish @vishhvak
150 Followers 2K Following AI Engineer @ BNY | MSAII at Carnegie Mellon University, Pittsburgh.
Reem Almutairi 🧠 @ReemNawaf
4K Followers 763 Following AI Engineer • creating AI that engage with your Brain
Simon Guo @simonguozirui
4K Followers 6K Following Beep Boop @thinkymachines | CS PhD student @Stanford | 🎓 @Berkeley_EECS
Marko Šišović @sisovicm
68 Followers 653 Following SWE @databricks by day pretraining research by night
GabboV @DVamanu61069
7 Followers 451 Following
curious @curiousity_much
1 Followers 519 Following
Naresh Neupane @naresh_neuupane
0 Followers 552 Following
AL @AngelLi14810900
2 Followers 524 Following
HZ @hongbinzhuang
20 Followers 367 Following
Efe Kaya @rockswe1
157 Followers 867 Following software engineer, working on ai infra & database internals MSCS '28 @tamu • 🇹🇷 18yo early college U.S. 🇺🇸 graduate
彡 Nᗩᖇᒪi 彡 @BGnarli
13K Followers 3K Following ┃Hello fellow! 💜┃ART CONDUCTOR┃Tech-Art┃Deep within latent space.┃#Art #digitalart #AIart ┃Air Thirsty
Arash Sadrieh @ArashSadrieh
114 Followers 378 Following Co-founder & Chief Scientist @NinjaTechAI | Building AI agents that work autonomously | Ex-Amazon | Sharing the raw journey of shipping AI
Luna Lovegood @NLPer_twitssc
42 Followers 2K Following Keep thinking about the meaning of being a PhD. | Opinions are my own What can I hold you with? Desperate sunsets and a lonely moon. — Jorge Luis Borges
Yifeng Ding @YifengDing_
972 Followers 3K Following CS PhD candidate @siebelschool. Research intern @AIatMeta. Building code LLMs and agents. Prev: @AmazonScience @GoogleResearch
Yu Lei @_OutofMemory_
716 Followers 3K Following PhD student @UTCompSci | Learn to understand ourselves and build intelligence.🤖🧠👁️
Alan Xing @alanxing
17K Followers 1K Following Co-founder/CEO of GenAI X, Inc. (GPT DAO) and Espeed Capital. @espeedcapital https://t.co/RWZUQhGswo
Lizhou Sha @mentisoasis
379 Followers 2K Following Postdoc @Princeton astrophysics | previously @UWMadison PhD '25, @TESSatMIT, @MIT '18 | planet hunter, programmer, violinist, sailor | he/him
Bhavya Chopra @BhavyaChopra1
1K Followers 818 Following CS PhD Student @UCBerkeley • previously at @Microsoft @ProseMSFT @Tableau @IIITDelhi • HCI+AI research for Data
Bassel Al Omari @BasselAlOmari_
279 Followers 222 Following research engineer @meta - ex. waabi, waterloo
Steven Dillmann @StevenDillmann
2K Followers 2K Following AI4Science PhD @Stanford @StanfordAILab Terminal-Bench-Science Lead @terminalbench Research Intern @allen_ai Prev. @Cambridge_Uni, @NASAJPL, @imperialcollege
Parsa Mirtaheri @parsamirta
345 Followers 1K Following PhD Student @UCSanDiego, Research Intern @Apple, IOI and ICPC medalist before AI won them
Wei Zhao @WeiMathRules
3 Followers 300 Following
return of the researc... @byebyescaling
3K Followers 2K Following ilya is right. bye bye scaling, back to empirical deep learning monkë | prev upenn, quant
Jean Baptiste Faddoul @faddoul_jean
12 Followers 714 Following
Hamed Zamani @HamedZamani
3K Followers 1K Following Assoc. Prof. @manningcics. Assoc. Director of the Center for Intelligent Information Retrieval (CIIR). Interested in IR, NLP & ML.
Simon Guo @simonguozirui
4K Followers 6K Following Beep Boop @thinkymachines | CS PhD student @Stanford | 🎓 @Berkeley_EECS
Rohan Jha @Robro612
430 Followers 526 Following PhD Student @jhuclsp | Prev: @mixedbreadai, @JinaAI_ Interested in IR & NLP
Kanat Özgen @ken_commodius
1 Followers 20 Following
Shangyang Min @shangyangmin
5 Followers 8 Following Founder https://t.co/M4n2GvXpGc | ML Content Douyin: lvbabydontcry | Prev. @BrownUniversity
Rikiya Takehi 武樋�... @rikiyatakehi_jp
1K Followers 73 Following PhD @MIT CSAIL | ex-@NVIDIA Research B.Eng @早稲田大学 English → @rikiyatakehi
Ao Qu @ao_qu18465
916 Followers 935 Following PhD at @MIT @mitidss @medialab | Building self-evolving AI & multisensory agents | Prev @ByteDanceSeed_
Huaibo Chen @hbchen21
26 Followers 106 Following PhD candidate @MIT | Intern @MITIBMLab | LLM agents / RL
elie @eliebakouch
25K Followers 5K Following research @PrimeIntellect (prev: @huggingface) anon feedback: https://t.co/JmMh7Sfvxd
Arman Cohan @armancohan
3K Followers 958 Following Assistant Professor of Computer Science @Yale Research Scientist @Ai2 LLM/NLP/AI Research
MIT CSAIL Alliances @csail_alliances
2K Followers 697 Following We connect companies to groundbreaking MIT CSAIL research. Our X account is paused. Engage with us here: https://t.co/HwFdID2U66
Scott Linderman @scott_linderman
6K Followers 950 Following Professor @Stanford Statistics and @StanfordBrain. Co-Founder of @EngramLab. AI, Neuroscience, Statistics. Posts are my own.
Jack Morris @jxmnop
54K Followers 1K Following research @engramlab // language models, information theory, science of AI
Jasper Lu @lu__jasper
1K Followers 280 Following continually learning. formerly taught models to design @figma, to see @nuro
Charlie O'Neill @oneill_c
22K Followers 1K Following The sea is the sea The old man is an old man The boy is a boy and the fish is a fish The sharks are all sharks no better and no worse
Alex Shaw @alexgshaw
3K Followers 760 Following Hacking on @terminalbench and @harborframework. Founding MTS @LaudeInstitute. Formerly Google. BYU alum.
Maksym Andriushchenko @maksym_andr
8K Followers 954 Following Principal investigator @ELLISInst_Tue & @MPI_IS, advisor @expsecai, mentor @MATSprogram. Past projects: AgentHarm, Claudini, PostTrainBench, Stolen Thoughts.
idan shenfeld @IdanShenfeld
2K Followers 198 Following PhD student @MIT | Prev student researcher @GoogleDeepMind
Taykhoom Dalal @Taykhoom_Dalal
67 Followers 250 Following CompBio PhD Candidate @MSKCancerCenter in @quaidmorris lab | Intern @MSFTResearch | @NSF GRFP Fellow | CS @UCLA
Derek Chong @dch
308 Followers 322 Following Building Shepherd w/ @StanfordNLP – git for agent execution.
CLS @ChengleiSi
7K Followers 4K Following Founding member @Recursive_SI & PhD @stanfordnlp | teaching language models to do research
Shayan Talaei @talaei_shayan
425 Followers 290 Following AI PhD @Stanford | LLM Research @RicursiveAI
Alexis Fox @alexisfox
437 Followers 124 Following Math & CS undergrad at Duke, → @MSFTResearch @MATSprogram
Braden Hancock @bradenjhancock
2K Followers 333 Following AI researcher, builder, backer. Research partner @LaudeInstitute + @LaudeVentures. Co-founder @SnorkelAI. Evals @Meta. Helping AI research become companies.
Ayden Zhang @AydenZhang0410
3 Followers 48 Following Building AI tools for real-world problems Currently helping international students understand lectures Youmi Lens @ Boston University
Kyunghyun Cho @kchonyc
89K Followers 2K Following a mediocre combination of a mediocre scientist and a mediocre advisor at @nyuniversity (@CILVRatNYU)
Isabelle Lee @wordscompute
266 Followers 274 Following ml/nlp phding @ usc, currently visiting harvard; training & interpretability & reasoning
Hyunjun Lee @ ICML 20... @Hyunjun_Lee1013
35 Followers 146 Following Undergrad Intern @ CLVR | ECE & Math @ SNU
Antonin Berthon @ntnnbrthn
3 Followers 39 Following
Ethan Mendes @EthanMendes3
165 Followers 544 Following Research Intern @MSFTResearch, ML PhD at @GeorgiaTech | Prev: Intern @CHAI_Berkeley
RK @kairosiann
132 Followers 503 Following pretraining world knowledge @yale ai robotic safety, startups, policy & governance meliorist at heart
Neil Kale @neilkale
538 Followers 372 Following founding mts @trajectorylabs + incoming phd @miteecs | prev. safety/alignment at scale ai, cmu
konsti. @KSturtzkopf
70 Followers 300 Following reinforcing abundance⋅ prev founding eng @joinhandshake ai
Minsik Oh @minsik_nlp
1K Followers 2K Following Research Scientist @StanfordAILab @StanfordNLP | CS @Stanford | persona, alignment & post-training, evals, representations
Li-Chun Lu @lichun_lu
14 Followers 101 Following Student researcher @ National Taiwan University (NTU) | Ex-Visiting Researcher @ UCLA PLUS lab | Research interest: human-AI Collaboration, AI Creativity, NLPTrends for United States
You might like
