Indraneil Paul @iNeil77
@ELLISForEurope PhD Candidate @UKPLab @TUDarmstadt ex. Research Intern @awscloud Working on RL + Code-Gen + Function Calling ineil77.dev Berlin, Germany Joined May 2018-
Tweets204
-
Followers163
-
Following297
-
Likes2K
Encouraging to see another western lab close the gap with the Chinese SOTA. Looks like important and interesting mid- and post-training work to be done in the coming releases!
Introducing Beam: a highly efficient agentic open model with 501B total parameters and 23B active. - Frontier reasoning efficiency - Advances the Western open frontier on coding & agentic tasks - Trained end-to-end from scratch Full weights release this month. Learn more about
@cloneofsimo Storage and load balancing for the flood of OpenJev bs that's about to hit
Hoping someone can explain to me what's going on here. The report says they trained multiple mixRL teachers, then used MOPD to combine those teachers into a student model. But..the team literally live-streamed their RL run, and then released model + report just one day after it finished. How does this timeline work out? I would assume the streamed RL run is just a final climb after MOPD, but the paper doesn't mention anything about a final climb. Maybe the livestreamed run is only for one of their mixRL teachers, and they did MOPD quickly in the 24 hours before launch?
MiMo-V2.6: The Hard Road to Scaling Up RL MiMo-V2.6 is very likely one of the largest single RL runs, by compute, that any open-source model team has undertaken to date. In an era when compute is brutally scarce, we still chose to dedicate a team of several dozen people to one
@menhguin Already the case with rolling venues like ARR, unfortunately
Twitter folk calling this "involution" in China meanwhile this has been a thing forever in India, where maybe a good 20% of grads from IITs eventually end up as instructors in JEE coaching classes or as DSA tutors 🤷♂️
Pretty insane PhD job market spillover problem in China. Shenzhen Middle School has over 100 teachers with a PhD, and 150 graduates of Tsinghua/PKU, 100+ alum of Harvard, Oxford, Cambridge You don’t need a PhD in Electrical Engineering from Tsinghua to teach middle school
@giffmana Also unclear how batch-level means would work in a multi-task RL setup
@dguido @trailofbits Would be way less fun though as the report would just be "here's how to sandbox like an adult" instead of an anthropomorphised slopvestigation 🤷♂️
@giffmana Interestingly, these entropy-as-memorization proxies are inconsistent across sub-fields. For instance, when training embedding models, we usually only retain the best memorized samples under the guise of “consistency filtering”.
Great to see the momentum behind rust now! Increasingly feels like the best language for CS101, especially in the age of AI.
Very cool to see: Rust is now a Tier-1 language at Microsoft
@maxidahl May have something to do with: arxiv.org/abs/2603.10145 Efficient tokenization needs more merges (and larger vocabularies), which can hinder learning.
@lu__jasper Not even sure the team publishing this trash paper is MAI! Just seems like some decisions were made in an isolated org to try and catapult a few careers. Also, Qwen is probs a year+ ahead in synthetic data generation and I don't think MAI corpus has parity with the Qwen3 corpus.
@dylan522p There isn't one on the planet west of Istanbul 🤷♂️
sharing a new long-form blog post: ai chip architectures it covers the leading chip architectures (nvidia, amd, tpus, trainium, cerebras, groq) across architecture, scaling (scale-up and scale-out), and software stacks. it helps build an intuition for the architectures and their trade-offs. jacobpeake.com/ai-chip-archit…
@arimorcos @giffmana I wonder if training on bad data annotated with reasons for its badness is as effective for VLMs as it is for LLMs (where it seems to beat overzealous filtering). Hard to retrofit into contrastive methods, but seems doable in captioning 🤔
@Dorialexander @pleiasfr @ynckdrt In our work on code RMs, we do observe that a 0.6B RM can be capable w.r.t. verifier accuracy evals, but lags behind a lot on downstream re-ranking and RL: openreview.net/forum?id=S2zGS…
In our work on code RMs, we do observe that a 0.6B RM can be capable w.r.t. verifier accuracy evals, but lags behind a lot on downstream re-ranking and RL: openreview.net/forum?id=S2zGS… Though looking at the download stats tells me people find it most handy anyway 😅
How small can a verifier be? Turns out 0.63M parameters, pretrained from scratch on synthetic data in about two minutes on one H100 can do as well as a 7B model on specific tasks. Check out the first of my small personal side projects, more to come! twozeros.eu/projects/verif…
@gabriberton Agree totally! It's not just about resources directly, but also about most confs devolving into a VC-style numbers game. Long-running projects with large-scale training are just a poor strategy in today's conf landscape.
Super cool long-horizon RL towards scientific replication by hill-climbing rubric rewards!
1/ Today, we introduce Faraday, a 27B-parameter AI Scientist that extends the capabilities of coding agents with a layer of scientific intuition. Trained via long-horizon RL, Faraday outperforms Claude Opus 4.8 and GPT-5.5 on the task of replicating research papers. 🧵
What a wonderful morning exercise!
Fable and I made a set of exercises for learning how PP schedules work: ezyang.github.io/pipeline-paral… -- unlike the other simulators, it asks you to build out the schedule block-by-block (double click to greedily finish the microbatch), subject to constraints
@ybnbxb @HeMuyu0327 There's a stop gradient on the denominator. So it still changes the gradient when pi_theta = pi_old compared to if the term was not there.
Thomas Zeng @tomzeng200
27 Followers 115 Following
Recuperado @honeydream41
11 Followers 1K Following
Varun Gangal @VarunGangal
2K Followers 9K Following LLM Evals / Model Behav Researcher @PatronusAI; prev @amazon AGI , PhD in AI via CMU adv. by @ehovy; @iitmcse '16 adv. by @ravi_iitm; history+geo+phil+angelinv!
Hossain Shaikh Saadi @saadi_hossain
53 Followers 387 Following Doctoral Researcher at @uni_mainz | Former MSc in Data Eng. & Analytics @TU_Muenchen, Graduate & Student Research Assistant @FraunhoferIIS, @CisLmu, @UKPLab
Nava @NavaRylove
26 Followers 215 Following
Andreas Waldis @AndreasWaldis
141 Followers 185 Following Robustness in Model Interpretability 🔎 PostDoc Tübingen University | previously PhD Student at UKP Lab, TU Darmstadt/Hochschule Luzern
leloy! @leloykun
8K Followers 6K Following MTS @mirendil • Math @ AdMU • NanoGPT speedrunner • Muon fan 🤍 • prev ML @ XPD • 2x IOI & 2x ICPC WF • https://t.co/nfO038itfn
Joey @djinnkering
36 Followers 3K Following
Bhupathi Guntupalli @bhupathi712
44 Followers 2K Following CS grad | Systems enthusiast Mega Bullish on life.
Nandan Thakur @beirmug
3K Followers 4K Following Postdoc @MSFTResearch | PhD @uwaterloo | RA @UKPLab l working on search and agents | (https://t.co/bxTCuxkJWZ, https://t.co/KYPd6PHDYd, @TREC_RAG and FreshStack)
Shivam Sharma @SharmaShivam027
9 Followers 223 Following
Wei Zhao @andyweizhao
394 Followers 645 Following Lecturer (assistant professor) in NLP @aberdeenuni.
shwetu (luca) @_shwetu
330 Followers 5K Following organic general intelligence | jack of all trades, master's from @NYUDataScience prev: Research @NYTimesRD @precog_iiitd; Manipal grad | he/him
Ashwin Vaswani @ashwin_vaswani
3K Followers 7K Following Research Scientist @GoogleDeepMind | Prev: @CarnegieMellon | @GoogleIndia | APPCAIR, @BITSPilaniGoa | @qtimlab, Harvard
Marcelo Formentão @marceloavf
9 Followers 3K Following
Louis @Louis9687221579
107 Followers 4K Following Mainline Economics | Idea page | ramblings of a schizo
Miguel @migdecamp
5 Followers 2K Following
Manish Gill @mgill25
4K Followers 8K Following Databases | Kubernetes | Compilers | EM @ClickHouseDB | https://t.co/12g8segwjE | https://t.co/worpODcnLt | Prev: AI Research @TomTom | Opinions my own
Richard @RMladek
1K Followers 1K Following chess international master, CTO @era_compute https://t.co/cTD9LqMPIu
Tyler evansville kate @tyler_kate69922
63 Followers 654 Following
Orion Weller @orionweller
3K Followers 1K Following Post-Training Gemini @GoogleDeepMind, PhD @jhuclsp
adamf @AdamForsted
46 Followers 621 Following
Aswin @aswingiridhar96
237 Followers 5K Following Tweets on AI and its impact. Opinions my own. He/ Him.
Théo Vincent @Theo_Vincent_
496 Followers 605 Following Intern @Meta, PhD student working on RL 🤖 @DFKI & @ias_tudarmstadt | Master MVA @ENS_ParisSaclay & ENPC 🎓
Xamu @Xamu15597
29 Followers 958 Following
AaronMichael🐉 @aaronmichael369
687 Followers 3K Following
Mengjie Zhao @mengjie_zhao
305 Followers 650 Following an NLP researcher @SBIntuitions @Softbank. Views my own. Previously researcher @Sony, phd @CisLmu. #nlproc
Boda Sadallah @bodasadallah
58 Followers 395 Following
Joe Stacey @_joestacey_
3K Followers 3K Following NLP postdoc at @SheffieldNLP Ex @Imperial_NLP PhD, @Apple AI/ML Scholar, @UCL MSc Robustness and uncertainty quantification
Chuanming @ChuanmingLiu
356 Followers 8K Following Ex-PhD student and alumni @sjtu1896 . Global citizen. Bootstrapping silicon-based life. Lifelong learning practitioner. Amateur triathlete and marathoner.
Charles Zhang @xiaojidan2011
166 Followers 5K Following
Neel Macwan @neelmacwantweet
123 Followers 1K Following Doctoral Student AI & Robotics & Mechanical | DAIM Research Group @ASU | 🦾 Robotics Grad at @MSNatASU | 🔱 Arizona State University
Rishu Kumar @rishdotuk
893 Followers 2K Following Long document understanding, Multilingual Evals and efficient models mainly, but other #NLProc applications in free time | vim enthusiast
dolphin @dolphin_lee_
1 Followers 357 Following
Shital Shah @sytelus
15K Followers 15K Following CEO and founder https://t.co/tTzeqS2PkF. Ex-Microsoft Research, Microsoft AI, Code infra lead Microsoft Phi models. If universe is an optimizer, what is its loss function?
Jesse Feld @JesseRSFeld
2K Followers 2K Following This platform is no longer in use, given the behaviour of its owner. Please go to https://t.co/1pfAMVcxUZ to find me on other platforms
HistoryFutureNow.com ... @tristanfischer
634 Followers 4K Following Author of https://t.co/55mhPeX51P, a website devoted to putting current events into a historical & future context. Centrist. RTs do NOT mean endorsement
Kexin Wang @KexinWang2049
296 Followers 123 Following A doctoral researcher at UKP @UKPLab @TUDarmstadt.
Walnut Jobs @WalnutJobs
530 Followers 59 Following A job board of @walnutai_ to repost hiring posts in high growth tech startups and giant tech. Help you locate key person and boost your career path❤️
Sander Dieleman @sedielem
71K Followers 2K Following Research Scientist at Google DeepMind (WaveNet, Nano Banana, Gemini Omni). I tweet about ML, music, generative models (personal account).
Geoffrey Irving @geoffreyirving
18K Followers 374 Following Cofounder and Chief Scientist at Resolution. Alignment will be solved, but not necessarily in time. Previously AISI, DeepMind, OpenAI, Google Brain, etc.
Peter O'Hearn @PeterOHearn12
2K Followers 149 Following Working on AI, code and reasoning. Researcher @AIatMeta & Prof @ucl. Separation logic, Incorrectness logic, Infer. Gödel Prize. Royal Society.
Nandan Thakur @beirmug
3K Followers 4K Following Postdoc @MSFTResearch | PhD @uwaterloo | RA @UKPLab l working on search and agents | (https://t.co/bxTCuxkJWZ, https://t.co/KYPd6PHDYd, @TREC_RAG and FreshStack)
Wei Zhao @andyweizhao
394 Followers 645 Following Lecturer (assistant professor) in NLP @aberdeenuni.
Maurice Kraus @ NeurI... @mkraus_io
75 Followers 77 Following PhD student @ AIML - TU Darmstadt. Passionate about finding ways to make complex models more transparent and understandable. #timeseries #XAI #research
Ashwin Vaswani @ashwin_vaswani
3K Followers 7K Following Research Scientist @GoogleDeepMind | Prev: @CarnegieMellon | @GoogleIndia | APPCAIR, @BITSPilaniGoa | @qtimlab, Harvard
Orion Weller @orionweller
3K Followers 1K Following Post-Training Gemini @GoogleDeepMind, PhD @jhuclsp
Baptiste Rozière @b_roziere
3K Followers 332 Following CodeGen lead at Mistral AI. Devstral/Codestral/Codestral Embed
Théo Vincent @Theo_Vincent_
496 Followers 605 Following Intern @Meta, PhD student working on RL 🤖 @DFKI & @ias_tudarmstadt | Master MVA @ENS_ParisSaclay & ENPC 🎓
Emperor @0xemperor
6K Followers 486 Following writing about crypto research/defi protocols crypto x ai || まだまだだね ||
Pranay Thangeda @pthangeda_
1K Followers 602 Following I like robots, and I spend my days teaching them to do (mostly) useful things at @amazon. Past @UofIllinois @NASA
Archiki Prasad @ArchikiPrasad
2K Followers 1K Following Research Scientist @GoogleDeepMind | Ph.D. from @unccs | @Apple AI/ML Scholar | Prev (intern): GDM, @AIatMeta (FAIR), @allenai_org
Dibya Ghosh @its_dibya
3K Followers 468 Following @AnthropicAI | Previously @ UCBerkeley @ Google Brain Montreal, @physical_int
Shital Shah @sytelus
15K Followers 15K Following CEO and founder https://t.co/tTzeqS2PkF. Ex-Microsoft Research, Microsoft AI, Code infra lead Microsoft Phi models. If universe is an optimizer, what is its loss function?
The AI Rev Hub - Jobs @AIRevHub_jobs
19 Followers 17 Following AI/ML LLMs are transforming the marketplace, and we're here to make sense of it all. We will be publishing jobs and tips to start working in an AI related job.
Asianometry @asianometry
19K Followers 347 Following My name is Jon and I run the Asianometry YouTube channel. You can email me at [email protected]
Math Cafe @Riazi_Cafe_en
55K Followers 113 Following The goal of this account is to encourage mathematical thinking.
Peiqin Lin @lpq29743
303 Followers 812 Following Algorithm Engineer @AlibabaGroup. Previously PhD at @ELLISforEurope @CisLmu @sardine_lab_it, interns at @HongKongPolyU, @TencentGlobal
Aaron Gokaslan @SkyLi0n
4K Followers 557 Following Creator of OpenWebText. @PyTorch/@pybind11 maintainer @Mozilla Rise25. Prev. @MBZUAI IFM Pretraining Infra Lead @Cornell (PhD) @FacebookAI and @BrownUniversity
Anmol Agarwal @anmolagarwal999
255 Followers 3K Following Grad Student @SCSatCMU @LTIatCMU Currently interested in problems in ML4Code, RL for LLMs, Alignment
Taco Cohen @TacoCohen
31K Followers 4K Following Slop janitor & post-trainologer at Meta / FAIR. Into codegen, RL, equivariance. Spent time at Qualcomm, Scyfer (acquired), UvA, Deepmind, OpenAI.
Ameya P. @AmyPrb
789 Followers 830 Following Postdoc at Tübingen (ELLIS & AI Center); Previously: @OxfordTVG Profile - https://t.co/To9NNR5Izc
Yifeng Ding @YifengDing_
988 Followers 3K Following CS PhD candidate @siebelschool. Research intern @AIatMeta. Building code LLMs and agents. Prev: @AmazonScience @GoogleResearch
Junlong Li @lockonlvange
530 Followers 334 Following PhD @hkust | MS/BS student @sjtu1896 | Ex Intern @deepseek_ai @MSFTResearch
Mathieu Blondel @mblondel_ml
9K Followers 466 Following Research scientist at Google DeepMind. Current research interests: differentiable programming, LLMs, Transformers.
GHTorrent @ghtorrent
580 Followers 1 Following A scalable, queriable, offline mirror of data offered through the Github REST API. Logo is from GitHub's Octicon font.
Nicholas Lourie @NickLourie
1K Followers 819 Following Better empirical methods for deep learning. PhD at @nyuniversity (@CILVRatNYU). Advised by @kchonyc and @hhexiy. Prev: @allen_ai. I build things. 🤖
Deep Learning For Cod... @DL4Code
1K Followers 34 Following Deep Learning For Code (DL4C) Workshop (ICLR'22, '23, '25, NeurIPS'25, and soon ICML'26)
Lucie Flek @lucie_nlp
4K Followers 4K Following #NLProc researcher, computer science prof @UniBonn Nothing new to see here in 2025. You will find my news via https://t.co/ms0AC3UfbJ #eXit
Justus-Jonas Erker �... @J_st_s
107 Followers 364 Following PhD Candidate @UKPLab | interested in representation learning, RAG & LLM Agents | MSc. AI from @MaastrichtU | prev. @DFKI #NLProc Supervisor: @Nils_Reimers
Not Parth Kolekar @parthkolekar
169 Followers 409 Following (he/him) ex-@Child, ex-@Infant | Building B2B SaaS
Abhinav Shukla @Abhinav95_
822 Followers 1K Following Applied Scientist at Amazon Robotics, ex: General Robotics, Meta, PhD at Imperial College London, IIIT Hyderabad
Anmol Goel @anmgoel
685 Followers 2K Following AI Safety x Privacy @ELLISForEurope PhD @UKPLab, @TUDarmstadt and @UCPH_Research | Prev @iiit_hyderabad |
Furkan Şahinuç @Furkansahinuc
100 Followers 182 Following NLP Doctoral Researcher @UKPLab Tweets are in EN/TR
Anurag Ghosh @anuragxel
503 Followers 745 Following PhDing @CMU_Robotics. CS @iiit_hyderabad. ex-@MSFTResearch. Trying to bring Science Fiction to Real Life.
Nico Daheim @ndaheim_
258 Followers 462 Following @ELLISforEurope PhD student in NLP and ML at @UKPLab @TUDarmstadt and @ETH_en. Previously MSc. in Data Science @RWTH.
Aishik Mandal @jit_aishik_13
106 Followers 241 Following ELLIS PhD Candidate @ UKP, TU Darmstadt | UG @ IIT KGP | Dialog Systems, Multumodality, NLP | Kopite🔴 | Mariner🟢🔴 | Otaku
Ilia Kuznetsov @ilokuznetsov
71 Followers 137 Following Postdoc at @UKPLab @TUDarmstadt🌳 // peer review, intertextuality, linguistics, NLP applications // synthesizers 🎹Trends for United States
You might like
