Sid @sidgraph
Energy, Taste and Intelligence e/acc @lossfunk, @Basethesislabs, @GenesisAILabs basethesis.com Earth Joined September 2019-
Tweets767
-
Followers1K
-
Following2K
-
Likes20K
really well written blog on the emerging shift from optimization-limited to rollout-limited frontier RL. what stood out to me was the observation that as reasoning trajectories become longer and more heterogeneous GPU utilization and learner-generator synchronization increasingly dominate training efficiency. one underexplored question is whether capability jumps themselves can be detected through staleness statistics. if old trajectories suddenly become much less useful for learning that may indicate the model has discovered a qualitatively new reasoning strategy. in that sense replay-buffer value decay could become an observable signal of phase transitions in capability acquisition. if you're into async-RL/large-scale training systems this is a really worthwhile read.
New blog! Is frontier asynchronous RL solved? The blog covers Async RL theory and infrastructure, surveying 8 open-weight frontier labs for the algorithmic techniques and systems fixes to handle train-inference mismatch. Also answered: why do current methods still fail at high
The open-source protein ML space just got a massive upgrade. Phenomenal work by @anindyadeeps and @try_litefold on dropping the biggest protein data collection on Hugging Face
We have released the biggest protein data collection on Hugging Face, guys! We have been working on this for more than 3 weeks now, starting from curating the raw data, doing a lot of filtering, splitting the datasets, sharding them, and doing a lot of analysis. Everything is
Check this out 👇
open sourcing Marlin-2B 🐟 a tiny VLM to extract structured information from videos Marlin is finetuned for two questions devs want to ask in their videos: what is happening, and when? Best open model in its weight class, competitive with Gemini-2.5-flash at only 2B params 🧵
Damn — its soo over for open ai 🫡
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
@nileshtrivedi yes i mean it was very obvious because this exists; but amazing work anyways :) arxiv.org/pdf/2403.12014
The bitter lesson in 26 words: Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
🚨Excited to announce Agent-BRACE! LLM agents in long-horizon POMDPs either blow up their context with raw history or summarize it, discarding uncertainty by collapsing belief into a point estimate. Agent-BRACE decouples the agent into belief state + policy models, jointly trained via RL. Key takeaways: 1️⃣ 🎯The belief state model produces a structured approximation of the belief distribution as a set of atomic natural-language claims with ordinal verbalized certainty labels ranging from certain to unknown. The policy conditions on this compact belief rather than the full history. 2️⃣ 📈 Outperforms strong RL baselines on long-horizon partially observable embodied language environments while maintaining a near-constant context window independent of episode length. 3️⃣ 🔄 The learned belief becomes increasingly calibrated as evidence accumulates, and epistemic belief decreases over time: the proportion of claims that the agent has the strongest level of belief in grows from 21% → 52% over an episode. 👇🧵
@thinkymachines released interaction models! ☄️ ➡️ decoder-only transformer running over a single interleaved (in_k, out_k) stream of 200ms micro-turns across audio, video, and text. Encoder-free early fusion: dMel intensity-binned mel filterbanks + light embed for audio, 40×40 patches through an hMLP stem for video, token embed for text; all collapsed into a bag of embeddings per 200ms chunk and co-trained from scratch. ➡️ Audio out is a flow-matching head over mel frames. No Whisper, no separate TTS, no VAD, no turn boundaries, no dialog manager. 276B/12B-active interaction model stays in real-time; an async background model handles tool use and long-horizon reasoning over shared context, with partial results merged back into the live stream at moments appropriate to user state. KAME and MoshiRAG did this pattern for retrieval; TML generalizes it. Streaming sessions (upstreamed to SGLang) ship each 200ms chunk as a separate request and append to a persistent in-GPU KV sequence, eliminating per-turn realloc and metadata recompute. MoE kernels use gather+gemv instead of grouped GEMM for decode shapes. Bitwise trainer-sampler alignment via batch-invariant kernels from their Sep 2025 work: NVLS deterministic all-reduce on Blackwell, Split-KV attention with consistent accumulation order between prefill and decode (SM-aligned 4096-token left splits), under 5% e2e overhead. Moshi hits 200ms but is audio-only. Qwen3-Omni is ~234ms but uses a separate AuT encoder + sliding-window DiT, not encoder-free. AURA does visual proactivity but wraps ASR/TTS around a VideoLLM (text-out, half-duplex). gpt-realtime-2 and Gemini Live still rely on VAD turn detection.
The technical report includes our motivation, early evaluation results, and technical approach. thinkingmachines.ai/blog/interacti…
Amazing talk by @DrJimFan -- exciting times for Physical AGI! TL;DR VLA architecture is parameter-misallocated toward language and should be replaced by World Action Models → pretrained video diffusion models that jointly predict future world states and robot actions, instantiated by Dream Zero (a 14B model running real-time control at 7Hz with 2× generalization gains over VLAs, @SeonghyeonYe ). arxiv.org/pdf/2602.15922 His central data claim is that egocentric human video is the FSD-equivalent ambient data flywheel for robotics, and EgoScale (@ruijie_zheng12) demonstrates a near-perfect log-linear scaling law (ℒ(N) = a − b log N, R² = 0.9983) between 1K and 20K hours of pretraining data and downstream dexterity performance. arxiv.org/pdf/2602.16710 His central environment claim is that classical physics simulators will be replaced by neural simulators, and Dream Dojo (@ShenyuanGao) demonstrates this with 44K hours of human video pretraining, 10 FPS real-time interaction, and Pearson r = 0.995 policy-evaluation fidelity. arxiv.org/pdf/2602.06949 Significant gaps in the talk: - it does not address runtime semantics (skill installation, behavior consistency, run-update separation) - it does not address the model-exploitation failure mode of training policies against learned simulators or learned rewards. My running notes w/ Opus 4.7 👇 docs.google.com/document/d/e/2… @NVIDIAAI
I promise this will be the best 20 min you spend today! Robotics: Endgame, the sequel to my last year's Sequoia AI Ascent talk, "Physical Turing Test". I laid out the roadmap for solving Physical AGI as a simple parallel to the LLM success story. Be a good scientist, copy
Notes on Robotics' End Game: Nvidia's Jim Fan! - @sidgraph and Opus 4.7 youtu.be/3Y8aq_ofEVs?si… TL;DR Vision-Language-Action architecture is parameter-misallocated toward language and should be replaced by World Action Models → pretrained video diffusion models that jointly predict future world states and robot actions, instantiated by Dream Zero (a 14B model running real-time control at 7Hz with 2× generalization gains over VLAs). His central data claim is that egocentric human video is the FSD-equivalent ambient data flywheel for robotics, and EgoScale demonstrates a near-perfect log-linear scaling law (ℒ(N) = a − b log N, R² = 0.9983) between 1K and 20K hours of pretraining data and downstream dexterity performance. His central environment claim is that classical physics simulators will be replaced by neural simulators, and Dream Dojo demonstrates this with 44K hours of human video pretraining, 10 FPS real-time interaction, and Pearson r = 0.995 policy-evaluation fidelity. The framework is strongest on data (genuinely empirically grounded), partial on architecture (Dream Zero shows the substrate works but is GPT-2-stage, not GPT-3-stage), and weakest on the predictive claims about timeline and the "VLA is dead" rhetorical framing. Three significant gaps in the talk: it does not address runtime semantics (skill installation, behavior consistency, run-update separation), it does not address safety, and it does not address the model-exploitation failure mode of training policies against learned simulators or learned rewards. docs.google.com/document/d/e/2…
@yacineMTB check out SWE-WebDevBench🥵 x.com/sidgraph/statu…
Vibe Coding is not vibing? Agents perform <60% on SWE-WebDevBench 👀. Code-level benchmarks (HumanEval, SWE-bench, FeatBench) take the specification as given and grade the patch. Vibe coding inverts this: the user gives natural-language intent, the platform must do PM,
Annnddd its a house full 😄🥳 its soo fun to bring best researchers of blr together — conversations are sickk cool
We’re hosting an invitation only gathering of researchers primarily in AI (but not limited to) over dinner in Indiranagar BLR! Just thoughtful conversations over good food, about seminal papers, emerging fields, research of your group/ recent papers you’ve published 🫶 If you
Amazing weather to discuss science and computing 🫶
We’re hosting an invitation only gathering of researchers primarily in AI (but not limited to) over dinner in Indiranagar BLR! Just thoughtful conversations over good food, about seminal papers, emerging fields, research of your group/ recent papers you’ve published 🫶 If you
Michael Bronstein @mmbronstein
58K Followers 8K Following #DeepMind Professor of #AI @UniofOxford / Director #AITHYRA / Chief Scientist @proximabio / https://t.co/kZpGpDAw4t (opinions are mine) 🤖🧪🧬🎶🐎
Chaitanya K. Joshi @chaitjo
10K Followers 1K Following AI researcher excited about biomolecule design 🧬 Postdoc @Stanford @RDasLab PhD student @Cambridge_Uni Prev. FAIR @AIatMeta @PrescientDesign @MRC_LMB
Ashish @v_ashish29
861 Followers 1K Following MSML @CarnegieMellon | Researcher @Cohere_Labs. Ex @IIScBangalore @UCF @NUSingapore
Francesco Di Giovanni @Francesco_dgv
2K Followers 191 Following I used to be a physicist / Riemannian geometer / GNN disciple; now I am working on generative models for drug discovery @RecursionPharma
Yash @yashgyy
2K Followers 2K Following How wonderful it is to have free will, laugh, talk, go anywhere, and to love with an unbounded capacity CSPhD Architecture@IITRKE | Mtech AI IITJ
Anna Brown @KiaNeu6
811 Followers 883 Following Commentary on The Worlds I See Ex-academic Disclaimer: If you identify with your ideas and cannot dissociate yourself from them, follow me at your own peril.
S V @open_parens
18 Followers 299 Following Love science, particuarly CS, Math. Previously: Ads TL, Google NYC, Microsoft Research, CSE @IITKgp
Nishant @nishantchandna_
22 Followers 559 Following DARPA Triage Challenge | Robotics | IROS 2025 Workshop | UAVs
pratham bhatnagar @prathamqq
1K Followers 3K Following vibe coding @deepworkai | https://t.co/Ft3ucgB5xz | 🦙/acc | 🍄 | trying to do all the hard things | @ns
pranay5255 @pranay5255
940 Followers 2K Following Reward hacking @lossfunk | ex @Kernel0x @Coinswitch
Abishek Suresh @abishek0suresh
2 Followers 37 Following
Jeanne @prompterminal
2K Followers 4K Following
Anirudh Balaji @AnirudhBalaji06
1 Followers 77 Following
Rtr @rtcoms
276 Followers 2K Following
amv @aryanmadhaverma
1K Followers 2K Following working with transformers and actuators before they replace me | prev: software @launchdarkly, ai @gethouseware, electr & math @bitspilaniindia
Rahul Sanghi @RahulSanghi1
5K Followers 1K Following Investing in India's most epic people and stories | Tigerfeathers | @nextupindia | ⛵
TomTom @TommyraNuccix
248 Followers 2K Following lol = Drowning Man. *lol* = Drowning Cheerleader. Like if you get it.
Kawin Ethayarajh @ethayarajh
5K Followers 1K Following Assistant Professor of Applied AI @ChicagoBooth @UChicago working on behavioral machine learning. PhD @StanfordAILab @stanfordnlp.
poorvivijay @PoorviVijay
499 Followers 1K Following SaaS investing at Elevation Capital | Harvard Business School | Ex-Adobe | Ex-Amazon | IIT G
Abhishek Eswaran @AbhishekEswaran
398 Followers 602 Following co-founder https://t.co/zzPSqjmJsj YCW26
Abhijeet Pranav Mishr... @aw__shucks
125 Followers 2K Following navigating b2b Saas || Ex: Bain & Co
Rishabh Rathod @rishabhrathod01
193 Followers 679 Following 28 | Frontend dev @prophecy_io | i usually raise complaints here
Shishira @Shishira97
25 Followers 6K Following
Surya A. Moitra @surya19m
308 Followers 4K Following here to bReak fRee,to rEdiscover mYself and perhaps to be 'coMforTably nUmB'..in parallel life- trying to make the world better with AI/ML
Vinayaks.eth @sharmavinayak24
1K Followers 6K Following Co-Founder NuvyLabs | Passionate Programmer | Cryptography | AI / ML |PolyMath | Technophile | Crypto Enthusiast | Engineer |
Div @hurtbadly2
109 Followers 2K Following Scientific computing / Ostracizing with Julia Cryo-ET, HEP, HPC 🍎🟣🍏
Sarthak Arora @sarthakarora128
19 Followers 111 Following
Akhilender Reddy @akhil_reddy_t
8 Followers 155 Following
Suhail Khan @_lazypoet_
3K Followers 4K Following Building robots to make fulfilment faster. gambling magic internet money.
Sudiksha Chindula @sudikshaoffl
4 Followers 282 Following
kushaL @_kyushaL
18 Followers 369 Following 1/7 time photographer . F1. cricket. Mech. Movies/Series.
Telecomblogs #Telecom... @telecomblogs
410 Followers 807 Following Past-Present-Future of Next Generation of Telecom. 5G/6G. Data-AI/ML & Edge Engineering.
Rakesh @followrakesh24
21 Followers 1K Following Senior ML Engineer ML systems in production Turning complexity into clear mental models Sleeps when tokens run out | Tennis
λux @novasarc01
22K Followers 3K Following tensor shepherd in a non-euclidean pasture | grazing on cuda cores
parichay_wtf @parichay_gg
255 Followers 1K Following love low level engineering, computer science and ai
Wasim Madha @wasim_madha
231 Followers 666 Following Founding Research Engineer at https://t.co/GYl8kPfOuJ | Making AI listen, speak & adapt.
shashwat @shashwatvalid
231 Followers 413 Following design, oss & hardware, doing side-quests @hackclub
François Chollet @fchollet
693K Followers 826 Following Co-founder @ndea. Co-founder @arcprize. Creator of Keras and ARC-AGI. Author of 'Deep Learning with Python'.
Michael Bronstein @mmbronstein
58K Followers 8K Following #DeepMind Professor of #AI @UniofOxford / Director #AITHYRA / Chief Scientist @proximabio / https://t.co/kZpGpDAw4t (opinions are mine) 🤖🧪🧬🎶🐎
Peyman Milanfar @docmilanfar
113K Followers 570 Following Distinguished Scientist at Google. National Academy of Engineering. Computational Imaging ∩ AI. Posts are personal opinions
Sebastian Raschka @rasbt
460K Followers 1K Following ML/AI research engineer. Ex stats professor. Author of "Build a Large Language Model From Scratch" (https://t.co/O8LAAMRzzW) & reasoning (https://t.co/5TueQKx2Fk)
Learning on Graphs Co... @LogConference
11K Followers 770 Following LoG is a new annual research conference that covers areas broadly related to machine learning on graphs and geometry, with a special focus on review quality.
Jeremy Howard @jeremyphoward
314K Followers 7K Following 🇦🇺 Co-founder: @AnswerDotAI/@FastDotAI ; Prev: Professor@UQ; @kaggle founding president; founder @fastmail/@enlitic/… https://t.co/16UBFTX7mo
elvis @omarsar0
307K Followers 854 Following Building self-improving AI @dair_ai • Prev: Meta AI | PhD • Learn about AI Agents for FREE here: https://t.co/P5SA9u54xO
Gabriel Peyré @gabrielpeyre
100K Followers 448 Following @CNRS researcher at @ENS_ULM. One tweet a day on computational mathematics.
Santiago @svpino
452K Followers 563 Following Computer scientist. I teach hard-core AI/ML Engineering at https://t.co/THCAAZcBMu. YouTube: https://t.co/pROi08OZYJ
Sergey Levine @svlevine
129K Followers 143 Following Associate Professor at UC Berkeley Co-founder, Physical Intelligence
Sanyam Bhutani @bhutanisanyam1
42K Followers 1K Following 👨💻 Working on llama models @AIatMeta | Previously: @h2oai, @weights_biases 🎙 Podcast @ctdsshow 👨🎓 Fellow @fastdotai 🎲 Grandmaster @Kaggle
Google DeepMind @GoogleDeepMind
1.4M Followers 279 Following The engine room of @Google. Building AI safely and responsibly to solve the world’s most complex problems. Join us: https://t.co/jUHQA27iBL
Kosta Derpanis (sabba... @CSProfKGD
81K Followers 199 Following #CS Assoc Prof @YorkUniversity, #ComputerVision Researcher, @VectorInst Affiliate, @UofT status only, @ELLISforEurope Member #CVPR2026 #ECCV2026 Publicity Chair
Michael Galkin @michael_galkin
8K Followers 346 Following Staff Research Scientist @GoogleAI. Prev: @Intel, Postdoc @Mila_Quebec & McGill. Graph Learning & LLMs. Grandmaster of 80's music (according to Spotify)
Jim Fan @DrJimFan
440K Followers 3K Following NVIDIA Director of Robotics & Distinguished Scientist. Co-Lead of GEAR lab. Solving Physical AGI, one motor at a time. Stanford Ph.D. OpenAI's 1st intern.
Radek Osmulski @radekosmulski
30K Followers 614 Following LLMs and retrieval by day and other genres of AI when I get the chance 🧪 Senior AI Eng @NVIDIAAI 🏫 @fastdotai trained DL Eng 📝 https://t.co/By87iXx5Pu
John Arnold @johnarnold
127K Followers 472 Following Co-chair of Arnold Ventures. Reality is more nuanced than the headline.
pranay5255 @pranay5255
940 Followers 2K Following Reward hacking @lossfunk | ex @Kernel0x @Coinswitch
Armin Parchami @ArminPCM
2K Followers 6K Following 👑✌️ Javid Shah ✌️👑 PhD | Sr. Director - Research Engineering @ Snorkel AI RL, Eval, & Benchmarking
Shruti Rajagopalan @srajagopalan
27K Followers 2K Following Economist @Mercatus Center @GeorgeMasonU. Fellow @nyulaw. Emergent Ventures India. Host @ideasofindia Podcast. Constitutional Economics & Public Choice. Dogs.
fleet @fleet_ai
2K Followers 4 Following Build better agents with better environments. Backed by Sequoia Capital, Menlo Ventures, and SVA.
Krish Gupta @krishgupta72
749 Followers 2K Following Kernel maxxing @ https://t.co/yMaj5SargB, GSoC'25 @gnuradio, GSoC'26 @llvmorg
Vinci @VinciPhysics
755 Followers 48 Following A frontier lab building the foundation model for the physical world, deployed in production on flagship engineering programs. Built by @hardikk13 + @saucentoss
Pixxel @PixxelSpace
36K Followers 8 Following Operating the world's highest-resolution commercial hyperspectral constellation of six satellites to deliver sharper, richer insights for a healthier planet! 🚀
Garth Sheldon-Coulson @garthsc
1K Followers 133 Following CEO @_panthalassa. Go where the energy is.
Panthalassa @_panthalassa
5K Followers 4 Following Building a planetary-scale energy platform in the middle of the ocean.
Kunvar Thaman @__kunvar__
3K Followers 904 Following Taking apart neural networks and putting them back together for a living. prev @si_pbc and @Akamai
Stephan Hoyer @shoyer
9K Followers 766 Following AI for science at @PeriodicLabs. Formerly, building AI climate models at Google. I also contribute to the scientific Python ecosystem (Xarray, NumPy, JAX).
Swayam Singh @swayaminsync
2K Followers 2K Following देखा एक ख्वाब तो ये सिलसिले हुए ✨ | @MSFTResearch | OSS Maintainer
Standard Intelligence @si_pbc
10K Followers 0 Following
Liam Fedus @LiamFedus
35K Followers 1K Following Building industrial-scale science at @periodiclabs Past: VP of Post-Training @OpenAI; Google Brain
λux @novasarc01
22K Followers 3K Following tensor shepherd in a non-euclidean pasture | grazing on cuda cores
elie @eliebakouch
17K Followers 4K Following training llm @PrimeIntellect (prev: @huggingface) anon feedback: https://t.co/JmMh7Sg3mL
Riya Bisht @b1shtream
1K Followers 5K Following accelerating biology using AI/computing @join_ef, neuromorphic @CeNSEatIISc, VC @Lvlupvc, @iiscbangalore,prev @Vicharak_In @CERN @BerkeleyLab, @ucberkeley
Ineffable Intelligenc... @IneffableLabs
7K Followers 1 Following Making first contact with superintelligence.
Vyom 👾 @HelloVyom
7K Followers 699 Following SDE Intern @amazon • Ex @samsung R&D • @jpmorgan CFG 2025 Winner • @amazon HackonS5 , Code With @cisco National Finalist • 30k+ @linkedin
Odyssey @odysseyml
32K Followers 12 Following We’re an AI lab pioneering world models—AI to understand and simulate the world.
Aditya Shrivastava @aditshri_
1K Followers 182 Following building something new @southpkcommons • nostalgia merchant @ https://t.co/TXm0Q6yYgr
Suryansh Shrivastava @Ro_leta_hu
6 Followers 274 Following SWE @Microsoft | Ex Research @Adobe, @IITD, @UNSW | Alumni @IIT Guwahati
Varul Srivastava @VarulSrivastava
139 Followers 708 Following Avg. at cooking and tennis. Curr @GoogleDeepMind prev. @Oracle @MSFTResearch @mll_iiith
Kushagra Vaish @kvaish_dev
732 Followers 564 Following SWE turned AI… something. Co-founder, https://t.co/17EKuLTWUc. he/him/his. Currently at Proximal
Pivotal Research @pivotal_org
833 Followers 136 Following Reducing global catastrophic risks from emerging technologies. https://t.co/ChghpGse8T – Apply by June 2nd
Siddhant Dubey @SiddhantD06
57 Followers 252 Following Early Stage Investor at General Catalyst https://t.co/ggYeye8E4J | Prev ML Engineer | Co-founder Aahar | IIT Delhi
Rahul Chhabra @rahulchhabra07
8K Followers 8K Following ceo @sabi. make something wonderful. taste is the bottleneck.
suhas motwani @MotwaniSuhas
23K Followers 2K Following building (DM if you’re exploring) / curating experiences @theproductfolks
Sanctuary 𖣐 @sanctuaryparc
1K Followers 8 Following a new habitat for founders building what's next | BLR
Paul Finney @paulfinneyx
6K Followers 4K Following Founder & Chief Tastemaker @spacekayak | Building @sanctuaryparc, a creative accelerator | BLR ↔ SF
Mingchen Zhuge @MingchenZhuge
2K Followers 874 Following Founding member @recursive_si; PhD of @SchmidhuberAI; Ex @Meta @Microsoft; Build @MetaGPT_, https://t.co/wnoau7ZPeE, https://t.co/OufDJP5DTO, https://t.co/tWu64hYA3T, ICLR RSI 2026, etc.













































