←
Jonathan Halcrow983 posts
Jonathan Halcrow avatar

Jonathan Halcrow

@jhalcrow

Feeding LLMs

◷ Joined February 2008▧ 49 Photos and videos
Vahab Mirrokni avatar
⟳ Jonathan Halcrow retweeted
Vahab Mirrokni@mirrokni·Sep 22

I'm happy to share that we have completed a proof of the Most Informative Boolean Function conjecture (a.k.a. Courtade–Kumar conjecture) in full generality, resolving a longstanding central open problem at the intersection of information theory and Boolean function analysis (arxiv.org/abs/2609.24931). We have also successfully verified analytic parts of the proof in Lean. This result emerged from an extensive human–AI collaboration with input from multiple AI models developed through a close partnership between our group at Google with Amin, Chandra, and Zijie at CHUK; See section 4 for details on our human-AI collaboration. The Courtade–Kumar conjecture was among the problems we set out to tackle in our AI-acceleration paper (arXiv:2602.03837). In this effort, we made substantial use of the Stellar Colosseum harness (arXiv:2609.15983) and a range of external and internal Gemini models. Background and Significance of CK Conjecture: - Posed by Thomas Courtade and P. R. Kumar ("Which Boolean Functions Maximize Mutual Information on Noisy Inputs?"), the conjecture posits that for a uniform input X in {-1, 1}^n passed through BSC(p) to produce Y, the mutual information I(f(X); Y) for any Boolean function f is maximized by coordinate projections f(x) = x_i, attaining the sharp upper bound 1 - h(p). - The conjecture belongs to a broader family of extremal problems that have attracted sustained attention. It has helped motivate parallel lines of work in information theory, theoretical computer science, Boolean function analysis, and discrete isoperimetry. - Kindler, O’Donnell, and Witmer document the considerable interest it has generated, and Yu and Tan’s 2022 monograph describes it as “one of the most important open problems in information theory” [28, Section 9.1.2]. - At a structural level, the conjecture is connected to foundational isoperimetric and functional inequalities, including Talagrand’s isoperimetric inequality and Bobkov’s inequality, as well as related conjectures on their optimal forms. In some regimes, it also recovers Harper's classical edge-isoperimetric inequality on the discrete cube. Last week, we learned that Ky and Tran had independently obtained a very different proof of the conjecture. We coordinated with them to release the two works separately citing the other. The two approaches offer substantially different approaches (arxiv.org/abs/2609.24184). Congratulations to Ky and Tran  as well on this great achievement. Stellar Colosseum is available externally as the long proof pattern of /teamwork in AntiGravity (antigravity.google/blog/team…). We plan to move the math x human collaboration features to AntiGravity and I am excited to provide this as a math collaboration platfom externally. Thanks for my wonderful coauthors Zijie Chen, Amin Gohari, @AdelJavanmard, Honghao Lin, Chandra Nair David Woodruff Link to paper: arxiv.org/abs/2609.24931

Vahab Mirrokni avatar
⟳ Jonathan Halcrow retweeted
Vahab Mirrokni@mirrokni·Aug 27

I’m proud to share our latest milestone. Following our Google I/O demo, we’ve pushed our Antigravity agent, /teamwork (powered by Gemini) from a helpful assistant into a true research partner from CS theory to high-performance coding and hardware simulation. Here is what we’ve been building 👇 antigravity.google/blog/team… 🔹 Designed a cycle-level RISC-V CPU simulator with minimum human intervention that successfully boots an OS to shell (< 1% average error vs. golden reference)💻⚙️ 🔹 Math & TCS (via Long-proof Pattern): We hit 71% on TCSBench and solved multiple open problems—including sparse convex optimization (JMLR), subspace approximation (FOCS), provable LLM quantization, vector embeddings, and Prefix-Matrix Factorizations. We also generated a simplified construction of Knuth’s Cycles Conjecture (verified in Lean) and a reproof of the Erdős Unit Distance Problem. 🧮🧠 🔹 Open Source: /teamwork writes production code. We landed performance optimizations directly into major C++ libraries, including ParlayHash and up to a 32× speedup in Eigen. ⚡🚀 What excites me isn't just the benchmarks—it's our shift from Thinking Graphs to Graph Engineering. We’ve moved past rigid prompt engineering and sequential flows. /teamwork treats long-horizon research as a dynamic execution graph: decoupling agent specifications from orchestration logic, running parallel tournament synthesis trees, and dynamically adjusting agent topologies at runtime as complex problems unfold. 🕸️🤖 We’re releasing several new harness patterns today to solve a variety of problems (iterative coding, document review, Long Proof, and Aletheia). I'm thrilled our team can share these agents via AGY with our users. Stay tuned for more! Read the full technical deep-dive on the blog here: antigravity.google/blog/team… #AI #Gemini #AI4Systems #AI4Math #AI4Science #ComputerArchitecture #Systems #FormalVerification #OpenSource

Sundar Pichai avatar
⟳ Jonathan Halcrow retweeted
Sundar Pichai@sundarpichai·15 Oct 2025

An exciting milestone for AI in science: Our C2S-Scale 27B foundation model, built with @Yale and based on Gemma, generated a novel hypothesis about cancer cellular behavior, which scientists experimentally validated in living cells.  With more preclinical and clinical tests, this discovery may reveal a promising new pathway for developing therapies to fight cancer.

Bahare Fatemi avatar
⟳ Jonathan Halcrow retweeted
Bahare Fatemi@BahareFatemi·27 Feb 2025

Think your LLM is smart? Try Big-Bench EXTRA Hard. The real test. #Reasoning #Benchmark #LLM

Is BIG-Bench Hard too easy for your LLM? We just unleashed BIG-Bench EXTRA Hard (BBEH)! 😈 Every task, harder! Every model, humbled! (Poem Credit: Gemini 2.0 Flash) Massive headroom for progress across various areas in general reasoning 🤯

Post attachment
Bahare Fatemi avatar
⟳ Jonathan Halcrow retweeted
Bahare Fatemi@BahareFatemi·14 Jun 2024

Think your LLM is a time wizard? 🧙 One might think they answer temporal questions through reasoning, but under the hood they may solely rely on their parametric knowledge 🤯 Our new benchmark, ToT, is specifically designed to measure the temporal reasoning capabilities of LLMs.

Post attachment
Anton Tsitsulin avatar
⟳ Jonathan Halcrow retweeted
Anton Tsitsulin@graph_·30 May 2024

🚨 Transformer theory alert 🚨 What algorithms can transformers execute efficiently? Our preprint sheds some light on reasoning capabilities of transformers, now in 𝒓𝒆𝒂𝒍𝒊𝒔𝒕𝒊𝒄 𝒑𝒂𝒓𝒂𝒎𝒆𝒕𝒆𝒓 𝒓𝒆𝒈𝒊𝒎𝒆𝒔. Paper: arxiv.org/abs/2405.18512 More in thread! 1/8

Post attachment
Bahare Fatemi avatar
⟳ Jonathan Halcrow retweeted
Bahare Fatemi@BahareFatemi·9 Feb 2024

Talk Like a Graph or Let Your Graph Do the Talking 🧠🕸️ New paper alert 🚨: we show how GNNs can effectively encode graph-structured data for LLMs: arxiv.org/abs/2402.05862

𝐋𝐞𝐭 𝐘𝐨𝐮𝐫 𝐆𝐫𝐚𝐩𝐡 𝐃𝐨 𝐭𝐡𝐞 𝐓𝐚𝐥𝐤𝐢𝐧𝐠: 𝐄𝐧𝐜𝐨𝐝𝐢𝐧𝐠 𝐒𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐃𝐚𝐭𝐚 𝐟𝐨𝐫 𝐋𝐋𝐌𝐬 Don't know what to do with your graphs in 2024? Shove them to an LLM, of course, and let LLM figure out what to do! arxiv.org/abs/2402.05862 Short thread (1/5):

Post attachment
Anton Tsitsulin avatar
⟳ Jonathan Halcrow retweeted
Anton Tsitsulin@graph_·9 Feb 2024

𝐋𝐞𝐭 𝐘𝐨𝐮𝐫 𝐆𝐫𝐚𝐩𝐡 𝐃𝐨 𝐭𝐡𝐞 𝐓𝐚𝐥𝐤𝐢𝐧𝐠: 𝐄𝐧𝐜𝐨𝐝𝐢𝐧𝐠 𝐒𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐃𝐚𝐭𝐚 𝐟𝐨𝐫 𝐋𝐋𝐌𝐬 Don't know what to do with your graphs in 2024? Shove them to an LLM, of course, and let LLM figure out what to do! arxiv.org/abs/2402.05862 Short thread (1/5):

Post attachment
Bryan Perozzi avatar
⟳ Jonathan Halcrow retweeted
Bryan Perozzi@phanein·30 Aug 2023

FYI: we're running a Kaggle contest with GNNs!

Kaggle@kaggle29 Aug 2023

📣 Competition launch alert! Google - Fast or Slow? Predict AI Model Runtime hosted by @Google Research. 🎯 predict the runtime of graphs and configurations in the test dataset 💰 $50,000 prize pool ⏰ November 10, 2023 (entry deadline) goo.gle/3qKJXg7

Post attachment
Anton Tsitsulin avatar
⟳ Jonathan Halcrow retweeted
Anton Tsitsulin@graph_·30 Jul 2023

Check out latest and greatest in data selection with graphs! The paper features: 🤯 A hardness theorem for training data selection 🕸️ A neural graph approach to solve the problem 🌶️ Comparisons with 10+ SotA methods No graph? No problem, just build one 🏗️ — I promise it works!

Tackling Provably Hard Representative Selection via Graph Neural Networks Mehran Kazemi, Anton Tsitsulin, Hossein Esfandiari et al.. Action editor: Bamdev Mishra. openreview.net/forum?id=3Lzg… #graphs #graph #nodes

Anton Tsitsulin avatar
⟳ Jonathan Halcrow retweeted
Anton Tsitsulin@graph_·28 Jul 2023

Aloha! 🤙🌺🌴 Check out our 3 papers today at @TAGinDS! (11AM p.) Learning Large Graph Property Prediction via Graph Segment Training (11AM p.) UGSL: A Unified Framework for Benchmarking Graph Structure Learning (2.30PM oral + 4PM p.) Unsupervised Embedding Quality Evaluation