{"version":1,"generated":"2026-08-14","license":"CC BY 4.0","fields":["id","title","claim","field","date","added","lab","model","verification","autonomy","tags","humans","year_posed","sources","url"],"total":70,"count":70,"offset":0,"entries":[{"id":"2026-08-13-sop2-sop3","title":"SOP_2 and SOP_3 theories shown to coincide","claim":"The classes of SOP_2 and SOP_3 first-order theories are the same, answering a 2004 question of Džamonja and Shelah and collapsing the bottom of the SOP hierarchy to a single class.","field":"mathematics","date":"2026-08-13","added":"2026-08-14","lab":"Independent","model":"ChatGPT 5.6","verification":"author-verified","autonomy":"collaborative","tags":["model-theory","logic","classification-theory"],"humans":["Artem Chernikov"],"year_posed":2004,"sources":[{"label":"arXiv: SOP_2 = SOP_3","url":"https://arxiv.org/abs/2608.13291","kind":"research"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com/problem/sop-2-sop-3","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-08-13-sop2-sop3"},{"id":"2026-08-12-liquid-drop-minimizers","title":"Complete minimizer picture for Gamow's liquid drop model","claim":"Balls uniquely minimize the Gamow liquid drop energy at every volume up to a sharp threshold of about 3.51, and above that threshold no minimizer exists, closing a gap that partial results had narrowed from both sides without meeting.","field":"mathematics","date":"2026-08-12","added":"2026-08-14","lab":"Independent","model":"ChatGPT 5.6 Pro","verification":"author-verified","autonomy":"ai-led","tags":["calculus-of-variations","geometric-analysis","mathematical-physics"],"humans":["Otis Chodosh","Matilde Gianocca"],"sources":[{"label":"arXiv: No compromise in the liquid drop model","url":"https://arxiv.org/abs/2608.11517","kind":"research"},{"label":"arXiv: An improved nonexistence bound for the liquid drop model","url":"https://arxiv.org/abs/2608.09000","kind":"commentary"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com/problem/gamow-liquid-drop-minimizer-conjecture","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-08-12-liquid-drop-minimizers"},{"id":"2026-08-10-zeta-zeros-critical-line","title":"Proportion of zeta zeros on the critical line raised to 67.25%","claim":"An unconditional proof that at least 67.25% of the nontrivial zeros of the Riemann zeta function are simple and lie on the critical line, up from a previous record of about 41.6%, with a machine-checked Lean formalization.","field":"mathematics","date":"2026-08-10","added":"2026-08-11","lab":"Anthropic","model":"Claude (unreleased research version)","verification":"formal","autonomy":"ai-led","tags":["lean","number-theory","riemann-hypothesis","multi-agent"],"humans":["Jarred Sumner","Levent Alpöge","Ralph Furman","Eric Easley"],"sources":[{"label":"Anthropic: paper (PDF)","url":"https://www-cdn.anthropic.com/564f962e60643842f5fcb4a17c9dbc8f608f1c37.pdf","kind":"research"},{"label":"Lean 4 formalization (anthropics/zeta-23-lean)","url":"https://github.com/anthropics/zeta-23-lean","kind":"research"},{"label":"Session transcript (PDF)","url":"https://www-cdn.anthropic.com/8a0d1add3c637b858a9a181e98c40e9548c3f44f.pdf","kind":"announcement"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-08-10-zeta-zeros-critical-line"},{"id":"2026-08-08-petersen-coloring","title":"A 112-vertex counterexample to the Petersen coloring conjecture","claim":"An explicit simple bridgeless cubic graph on 112 vertices admits no Petersen coloring, and hence no normal 5-edge-coloring, refuting a conjecture of Jaeger from 1985 with machine-checked UNSAT certificates.","field":"mathematics","date":"2026-08-08","added":"2026-08-14","lab":"Independent","model":"Unnamed OpenAI model","verification":"formal","autonomy":"ai-assisted","tags":["graph-theory","counterexample","sat-solver","combinatorics"],"humans":["Bryce Putman"],"year_posed":1985,"sources":[{"label":"arXiv: A 112-Vertex Counterexample to the Petersen Coloring Conjecture","url":"https://arxiv.org/abs/2608.10012","kind":"research"},{"label":"Zenodo: reproducibility artifacts and checked DRAT certificates","url":"https://doi.org/10.5281/zenodo.21845291","kind":"research"},{"label":"Open Problem Garden: Petersen coloring conjecture","url":"https://www.openproblemgarden.org/op/petersen_coloring_conjecture","kind":"commentary"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com/problem/petersen-coloring-conjecture","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-08-08-petersen-coloring"},{"id":"2026-08-05-schiffer-conjecture","title":"Counterexamples to Schiffer's conjecture and the Pompeiu problem","claim":"Infinitely many planar domains that are not balls admit a Neumann eigenfunction of the Laplacian that is constant on the boundary, disproving Schiffer's conjecture and, through a classical equivalence, the 1929 Pompeiu problem.","field":"mathematics","date":"2026-08-05","added":"2026-08-14","lab":"Independent","model":"GPT-5.6, Claude Opus 4.8, Claude Fable 5","verification":"formal","autonomy":"ai-assisted","tags":["lean","spectral-geometry","counterexample","analysis"],"humans":["Gonzalo Cao-Labora","Jaume de Dios Pont"],"year_posed":1929,"sources":[{"label":"arXiv: Counterexamples to Schiffer's Conjecture","url":"https://arxiv.org/abs/2608.05114","kind":"research"},{"label":"Lean 4 formalization (jaumededios/Schiffer)","url":"https://github.com/jaumededios/Schiffer","kind":"research"},{"label":"Wikipedia: Pompeiu problem","url":"https://en.wikipedia.org/wiki/Pompeiu_problem","kind":"commentary"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com/problem/schiffer-conjecture","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-08-05-schiffer-conjecture"},{"id":"2026-08-05-sendov-conjecture","title":"Sendov's conjecture proved for every degree","claim":"For every complex polynomial of degree at least two whose zeros all lie in the closed unit disk, each zero has a critical point of the polynomial within distance one, closing a question open since 1959.","field":"mathematics","date":"2026-08-05","added":"2026-08-14","lab":"Independent","model":"GPT-5.6 Pro","verification":"formal","autonomy":"collaborative","tags":["lean","complex-analysis","polynomials","formalization"],"humans":["Lech Mazur"],"year_posed":1959,"sources":[{"label":"ProofAtlas: A Computer-Assisted Proof of Sendov's Conjecture (PDF)","url":"https://www.proofatlas.ai/papers/sendov-conjecture/SENDOV_CONJECTURE_PROOF_AUGUST_5_2026.pdf","kind":"research"},{"label":"ProofAtlas: Sendov conjecture formalization record","url":"https://proofatlas.ai/formalizations/sendov-conjecture/","kind":"research"},{"label":"Terence Tao: A digestion of the proof of Sendov's conjecture","url":"https://terrytao.wordpress.com/2026/08/12/a-digestion-of-the-proof-of-sendovs-conjecture/","kind":"commentary"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com/problem/sendov-s-conjecture","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-08-05-sendov-conjecture"},{"id":"2026-08-04-moore-bound","title":"Asymptotic degree-diameter problem resolved for fixed diameter","claim":"The maximum order of a graph with maximum degree d and diameter k satisfies n_k(d)/d^k tending to 1 as d grows, for every fixed k, resolving the asymptotic form of a question open since 1978, with a Lean 4 formalization.","field":"mathematics","date":"2026-08-04","added":"2026-08-11","lab":"Independent","model":"GPT-5.6 Pro","verification":"formal","autonomy":"ai-assisted","tags":["lean","graph-theory","combinatorics","extremal-combinatorics"],"humans":["Wouter Cames van Batenburg","Samuel Korsky"],"year_posed":1978,"sources":[{"label":"arXiv: Asymptotically attaining the Moore bound","url":"https://arxiv.org/abs/2608.03965","kind":"research"},{"label":"Lean 4 formalization (woutercvb/wewantmoore)","url":"https://github.com/woutercvb/wewantmoore","kind":"research"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-08-04-moore-bound"},{"id":"2026-08-01-astra-ten-advances","title":"Ten results in mathematics and theoretical computer science with Lean certificates","claim":"An unreleased OpenAI model produced ten results including the first explicit non-sofic group and a disproof of Connes's rigidity conjecture, each published with a machine-checkable Lean 4 proof.","field":"mathematics","date":"2026-08-01","added":"2026-08-06","lab":"OpenAI","model":"Astra","verification":"formal","autonomy":"ai-led","tags":["group-theory","lean","formalization","sphere-packing","quantum-complexity"],"year_posed":1999,"sources":[{"label":"Lean 4 certificates for all ten results","url":"https://github.com/openai/ten-proofs","kind":"research"},{"label":"OpenAI: Ten advances in mathematics and theoretical computer science","url":"https://openai.com/index/ten-advances-in-mathematics/","kind":"announcement"}],"url":"https://whataifound.org/finding/2026-08-01-astra-ten-advances"},{"id":"2026-07-29-sumset-difference-exponent","title":"Optimal exponent relating sumsets and difference sets determined","claim":"The least universal exponent c with |A+A|/|A| bounded by (|A-A|/|A|)^c for every finite integer set A is determined, settling the extremal question with a Lean 4 formalization.","field":"mathematics","date":"2026-07-29","added":"2026-08-11","lab":"Tencent Hunyuan","model":"Hy3 (Hyra research agent)","verification":"formal","autonomy":"ai-led","tags":["lean","additive-combinatorics","number-theory"],"humans":["Haowei Lin","Shanda Li"],"sources":[{"label":"arXiv: Settling the Optimal Exponent Relating Sumsets and Difference Sets","url":"https://arxiv.org/abs/2607.27199","kind":"research"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-07-29-sumset-difference-exponent"},{"id":"2026-07-28-kemeny-three-voters","title":"Kemeny rank aggregation shown NP-hard for three voters","claim":"Computing a Kemeny-optimal aggregate ranking is NP-hard when the input is exactly three complete rankings, closing the minimal open case left by hardness results for even voter counts.","field":"computer-science","date":"2026-07-28","added":"2026-08-11","lab":"Independent","model":"GPT-5.6 Sol Ultra, Claude Fable 5","verification":"formal","autonomy":"ai-led","tags":["lean","complexity-theory","computational-social-choice"],"humans":["Dominik Peters"],"year_posed":2001,"sources":[{"label":"arXiv: Kemeny Rank Aggregation is NP-Hard for Three Voters","url":"https://arxiv.org/abs/2607.25540","kind":"research"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-07-28-kemeny-three-voters"},{"id":"2026-07-27-feige-conjecture","title":"Feige's conjecture on sums of nonnegative random variables settled","claim":"For independent nonnegative random variables with mean at most 1 and sum S, the probability that S is below its mean plus one is at least 1/e, the sharp constant Feige conjectured in 2004; three independent proofs appeared within days.","field":"mathematics","date":"2026-07-27","added":"2026-08-11","lab":"Independent","model":"ChatGPT 5.6 Pro","verification":"formal","autonomy":"ai-led","tags":["lean","probability","concentration-inequalities"],"humans":["Weibo Fu","Yanjun Han","Guanyang Wang","Jun Yan","Peng Zhang","Zhengqing Zhou"],"year_posed":2004,"sources":[{"label":"arXiv: Sharp small-deviation inequalities for sums of independent nonnegative random variables","url":"https://arxiv.org/abs/2607.23980","kind":"research"},{"label":"Lean formalization (pengzhang91/Feige)","url":"https://github.com/pengzhang91/Feige","kind":"research"},{"label":"arXiv: On Feige's conjecture (Nie and Wei)","url":"https://arxiv.org/abs/2607.24528","kind":"research"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-07-27-feige-conjecture"},{"id":"2026-07-22-dinitz-garg-goemans","title":"Counterexample to the Dinitz–Garg–Goemans conjecture","claim":"An explicit seven-node network whose cheapest single-route (unsplittable) shipping costs more than its fractional cost even under the allowed capacity slack, disproving a conjecture from the late 1990s.","field":"computer-science","date":"2026-07-22","added":"2026-07-24","lab":"Independent","model":"GPT-5.6 Pro","verification":"claimed","autonomy":"ai-led","tags":["combinatorial-optimization","network-flow","counterexample","unsplittable-flow"],"humans":["Dmitry Rybin"],"year_posed":1999,"sources":[{"label":"Dmitry Rybin announcement (X)","url":"https://x.com/DmitryRybin1/status/2079907499545919968","kind":"announcement"},{"label":"Coverage: officechai","url":"https://officechai.com/ai/mathematician-says-gpt-5-6-disproved-the-30-year-old-dinitz-garg-goemans-conjecture-with-4-simple-prompts/","kind":"coverage"}],"url":"https://whataifound.org/finding/2026-07-22-dinitz-garg-goemans"},{"id":"2026-07-20-gaussian-moments","title":"Counterexamples to the Gaussian moments conjecture","claim":"Explicit low-degree polynomials disproving a 2017 conjecture that had been proposed as a route to proving the Jacobian conjecture, found days after that conjecture was itself disproved.","field":"mathematics","date":"2026-07-20","added":"2026-07-27","lab":"Independent","model":"GPT-5.6 Sol Pro + Claude Fable 5","verification":"author-verified","autonomy":"ai-led","tags":["algebra","polynomial-maps","counterexample","jacobian-conjecture"],"humans":["Christopher D. Long"],"year_posed":2017,"sources":[{"label":"arXiv: Small Counterexamples to the Gaussian Moments Conjecture","url":"https://arxiv.org/abs/2607.18186","kind":"research"},{"label":"Derksen, van den Essen, Zhao: The Gaussian Moments Conjecture and the Jacobian Conjecture","url":"https://arxiv.org/abs/1506.05192","kind":"research"}],"url":"https://whataifound.org/finding/2026-07-20-gaussian-moments"},{"id":"2026-07-20-gaussian-product-inequality","title":"Gaussian product inequality conjecture proved","claim":"For any centered Gaussian vector and positive exponents, the expectation of the product of absolute powers is at least the product of the individual expectations, settling a conjecture posed in 2007.","field":"mathematics","date":"2026-07-20","added":"2026-08-11","lab":"Independent","model":"ChatGPT 5.6 Sol","verification":"formal","autonomy":"ai-led","tags":["lean","probability","gaussian-inequalities"],"humans":["Frédéric Ouimet","Dylan Greaves"],"year_posed":2007,"sources":[{"label":"A proof of the strong Gaussian product inequality conjecture","url":"https://doi.org/10.13140/RG.2.2.17569.77923/1","kind":"research"},{"label":"Lean formalization (dylgre/gaussian-product-inequality)","url":"https://github.com/dylgre/gaussian-product-inequality","kind":"research"},{"label":"Public chat transcript","url":"https://chatgpt.com/share/6a5ea69b-1648-83e8-80b1-014ae0b1003c","kind":"announcement"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-07-20-gaussian-product-inequality"},{"id":"2026-07-20-kourovka-notebook","title":"Eight problems from the Kourovka Notebook solved and formalized in Lean","claim":"A formal reasoning agent developed the proof strategies for eight open problems in group theory without human guidance on which mathematical steps to take, and machine-checked every one in Lean 4.","field":"mathematics","date":"2026-07-20","added":"2026-08-06","lab":"Harmonic","model":"Aristotle","verification":"formal","autonomy":"autonomous","tags":["group-theory","lean","formalization","kourovka-notebook","autonomous-discovery"],"humans":["Wouter van Doorn","Elias Judin","Pietro Monticone","Daniel Morrison"],"year_posed":1969,"sources":[{"label":"arXiv: On Some Problems from the Kourovka Notebook","url":"https://arxiv.org/abs/2607.17477","kind":"research"},{"label":"Lean 4 formalizations of all eight solutions","url":"https://github.com/pitmonticone/Kourovka","kind":"research"}],"url":"https://whataifound.org/finding/2026-07-20-kourovka-notebook"},{"id":"2026-07-19-jacobian-conjecture","title":"Counterexample to the Jacobian conjecture in dimension three","claim":"An explicit polynomial map in three variables with constant Jacobian determinant −2 that is nevertheless not invertible, disproving a conjecture open since 1939.","field":"mathematics","date":"2026-07-19","added":"2026-07-20","lab":"Anthropic","model":"Claude Fable 5","verification":"formal","autonomy":"collaborative","tags":["algebraic-geometry","counterexample","polynomial-maps"],"humans":["Levent Alpöge"],"year_posed":1939,"sources":[{"label":"The Jacobian counterexample, explained","url":"https://jacobianfun.org/jacobian-explained","kind":"research"},{"label":"Alexis Gallagher: Jacobian Conjecture Disproved!","url":"https://alexisgallagher.com/posts/2026/jacobianfun/","kind":"commentary"},{"label":"Hacker News discussion","url":"https://news.ycombinator.com/item?id=48973869","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-07-19-jacobian-conjecture"},{"id":"2026-07-14-sabidussi-compatibility","title":"Sabidussi's compatibility conjecture proved","claim":"The edges of a finite connected multigraph carrying a closed eulerian trail can be partitioned into circuits so that no circuit contains two edges used consecutively in the trail, with a Lean 4 formalization.","field":"mathematics","date":"2026-07-14","added":"2026-08-11","lab":"Independent","model":"GPT-5.6 Pro, GPT-5.6 Sol","verification":"formal","autonomy":"collaborative","tags":["lean","graph-theory","combinatorics"],"humans":["Nikolay Ulyanov"],"sources":[{"label":"arXiv: Graph Puzzles III.1: A Proof of Sabidussi's Compatibility Conjecture","url":"https://arxiv.org/abs/2607.13225","kind":"research"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-07-14-sabidussi-compatibility"},{"id":"2026-07-14-zeroth-order-oracle-bound","title":"Near-quadratic lower bound for derivative-free convex optimization","claim":"A lower bound of order d²/log d on the number of exact function evaluations needed to minimize a convex Lipschitz function, closing a gap open since 1996 and showing a 1996 algorithm was essentially optimal.","field":"mathematics","date":"2026-07-14","added":"2026-07-27","lab":"Independent","model":"GPT-5.6 Sol Pro","verification":"formal","autonomy":"ai-led","tags":["optimization","complexity-theory","lower-bounds","lean"],"humans":["Phillip Kerger"],"year_posed":1996,"sources":[{"label":"arXiv: Closing the Oracle-Complexity Gap in Derivative-Free Convex Optimization","url":"https://arxiv.org/abs/2607.13335","kind":"research"},{"label":"Lean verification, prompt and chat logs","url":"https://github.com/PhillipKerger/zero-order-bounds-lean-verification","kind":"research"}],"url":"https://whataifound.org/finding/2026-07-14-zeroth-order-oracle-bound"},{"id":"2026-07-11-grothendieck-group-schemes","title":"Counterexample to Grothendieck's question on finite flat group schemes","claim":"A finite locally free group scheme of order four whose fourth power map is not trivial, answering a question Grothendieck raised in the 1960s; one model produced the construction and another formalised it in Lean.","field":"mathematics","date":"2026-07-11","added":"2026-07-29","lab":"Independent","model":"OpenAI Sol (construction); Claude Fable (Lean formalisation)","verification":"formal","autonomy":"collaborative","tags":["algebraic-geometry","group-schemes","counterexample","lean","formalization"],"humans":["Akhil Mathew","Kevin Buzzard"],"sources":[{"label":"Kevin Buzzard: Human mathematicians are being outcounterexampled","url":"https://xenaproject.wordpress.com/2026/07/20/human-mathematicians-are-being-outcounterexampled/","kind":"announcement"},{"label":"mathlib4 PR #41748: a finite free group scheme of order four not killed by four","url":"https://github.com/leanprover-community/mathlib4/pull/41748","kind":"research"}],"url":"https://whataifound.org/finding/2026-07-11-grothendieck-group-schemes"},{"id":"2026-07-10-cycle-double-cover","title":"Cycle double cover conjecture proved for all bridgeless multigraphs","claim":"Every finite bridgeless loopless multigraph has a multiset of cycles covering each edge exactly twice, resolving a conjecture open since 1973, with a machine-checked Lean proof of the unconditional statement.","field":"mathematics","date":"2026-07-10","added":"2026-08-14","lab":"OpenAI","model":"GPT-5.6 Sol Ultra","verification":"formal","autonomy":"ai-led","tags":["lean","graph-theory","combinatorics","multi-agent"],"year_posed":1973,"sources":[{"label":"OpenAI: A proof of the cycle double cover conjecture (PDF)","url":"https://cdn.openai.com/pdf/04d1d1e4-bc75-476a-97cf-49055cd98d31/cdc_proof.pdf","kind":"research"},{"label":"Lean 4 formalization (openai/cdc-lean)","url":"https://github.com/openai/cdc-lean","kind":"research"},{"label":"OpenAI: the prompt used for the proof (PDF)","url":"https://cdn.openai.com/pdf/04d1d1e4-bc75-476a-97cf-49055cd98d31/cdc_prompt.pdf","kind":"announcement"},{"label":"Ethan Knight announcement (X)","url":"https://x.com/__eknight__/status/2075643450196971805","kind":"announcement"},{"label":"arXiv: A proof of the cycle double cover conjecture by OpenAI: An exposition","url":"https://arxiv.org/abs/2607.16356","kind":"commentary"},{"label":"arXiv: Three-Bit Flows and Cycle Covers. Part I","url":"https://arxiv.org/abs/2607.14140","kind":"commentary"},{"label":"Scientific American: ChatGPT just proved another 50-year-old math conjecture","url":"https://www.scientificamerican.com/article/chatgpt-just-proved-another-50-year-old-math-conjecture/","kind":"coverage"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com/problem/cycle-double-cover-conjecture","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-07-10-cycle-double-cover"},{"id":"2026-06-25-herculaneum-scroll-read","title":"First Herculaneum scroll read end to end without unrolling it","claim":"PHerc. 1667 was virtually unwrapped and read in full from X-ray scans using learned ink detection: about 1.4 metres of papyrus across roughly 22 columns of Greek, identified as Philodemus, On Gods, Book 8.","field":"archaeology","date":"2026-06-25","added":"2026-07-29","lab":"Vesuvius Challenge","model":"Community-developed ink-detection neural networks","verification":"author-verified","autonomy":"ai-assisted","tags":["papyrology","virtual-unwrapping","ink-detection","ancient-texts"],"humans":["Brent Seales","Luke Farritor","Youssef Nader","Julian Schilliger"],"sources":[{"label":"Complete virtual unwrapping and reading of a rolled Herculaneum papyrus","url":"https://arxiv.org/abs/2606.29085","kind":"research"},{"label":"Vesuvius Challenge: An entire Herculaneum scroll has been read for the first time","url":"https://scrollprize.org/firstscroll","kind":"announcement"},{"label":"Vesuvius Challenge","url":"https://scrollprize.org/","kind":"announcement"}],"url":"https://whataifound.org/finding/2026-06-25-herculaneum-scroll-read"},{"id":"2026-06-17-kagome-superconductors","title":"Two kagome superconductors predicted by machine learning and confirmed in the lab","claim":"A screening pipeline that narrowed over 1.3 million candidate structures picked out YRu3B2 and LuRu3B2, which were then synthesized and measured to superconduct at 0.81 K and 0.95 K.","field":"materials","date":"2026-06-17","added":"2026-08-06","lab":"Aalto University / Rice University","model":"Unknown","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["superconductivity","kagome","high-throughput-screening","dft","experimental-confirmation"],"humans":["Rose Albu Mustaf","Päivi Törmä","Emilia Morosan","B. Andrei Bernevig","Miguel A. L. Marques"],"sources":[{"label":"arXiv: Machine Learning-Guided Discovery of Kagome Superconductors YRu3B2 and LuRu3B2","url":"https://arxiv.org/abs/2512.16945","kind":"research"},{"label":"ScienceDaily: AI just supercharged the race to find room temperature superconductors","url":"https://www.sciencedaily.com/releases/2026/07/260701205006.htm","kind":"coverage"}],"url":"https://whataifound.org/finding/2026-06-17-kagome-superconductors"},{"id":"2026-06-02-jamming-exponent-identity","title":"Identity for the critical exponents of jamming derived analytically","claim":"The relation a + b = 1 between critical exponents at the jamming transition, previously seen only numerically in the full replica-symmetry-breaking solution of hard spheres, is derived from the scaling equations and published after peer review.","field":"physics","date":"2026-06-02","added":"2026-08-11","lab":"Independent","model":"Claude Sonnet 4.6, Claude Opus 4.7","verification":"peer-reviewed","autonomy":"collaborative","tags":["statistical-physics","jamming","critical-exponents"],"humans":["Giorgio Parisi","Francesco Zamponi"],"year_posed":2014,"sources":[{"label":"arXiv: A proof of an identity for the critical exponents of jamming","url":"https://arxiv.org/abs/2606.03300","kind":"research"},{"label":"vibemathed.com registry entry","url":"https://vibemathed.com","kind":"commentary"}],"url":"https://whataifound.org/finding/2026-06-02-jamming-exponent-identity"},{"id":"2026-05-21-alphaproof-nexus","title":"Nine Erdős problems and 44 OEIS conjectures proved with machine-checked proofs","claim":"An agent pairing language models with the Lean proof assistant resolved 9 of 353 open Erdős problems and 44 of 492 open OEIS conjectures, with every proof mechanically verified.","field":"mathematics","date":"2026-05-21","added":"2026-07-27","lab":"Google DeepMind","model":"AlphaProof Nexus (LLM + Lean)","verification":"formal","autonomy":"search-scaffold","tags":["lean","automated-theorem-proving","erdos-problems","combinatorics"],"humans":["Pushmeet Kohli","Swarat Chaudhuri"],"sources":[{"label":"arXiv: Advancing Mathematics Research with AI-Driven Formal Proof Search","url":"https://arxiv.org/abs/2605.22763","kind":"research"},{"label":"All Lean proofs (DeepMind repository)","url":"https://github.com/google-deepmind/alphaproof-nexus-results","kind":"research"},{"label":"Tao's AI-contributions-to-Erdős-problems wiki","url":"https://github.com/teorth/erdosproblems/wiki/AI-contributions-to-Erd%C5%91s-problems","kind":"research"}],"url":"https://whataifound.org/finding/2026-05-21-alphaproof-nexus"},{"id":"2026-05-erdos-unit-distance","title":"Disproof of the Erdős unit-distance conjecture","claim":"A reasoning model produced the core construction disproving a 1946 conjecture in discrete geometry, finding point configurations with more unit-distance pairs than the conjecture permitted.","field":"mathematics","date":"2026-05-20","added":"2026-07-20","lab":"OpenAI","model":"GPT-5 series reasoning model","verification":"independent","autonomy":"ai-led","tags":["discrete-geometry","combinatorics","erdos"],"year_posed":1946,"sources":[{"label":"Remarks on the disproof of the unit distance conjecture (human-verified writeup)","url":"https://arxiv.org/abs/2605.20695","kind":"research"},{"label":"Sawin: An explicit lower bound for the unit distance problem","url":"https://arxiv.org/abs/2605.20579","kind":"research"},{"label":"Quanta: The AI Revolution in Math Has Arrived","url":"https://www.quantamagazine.org/the-ai-revolution-in-math-has-arrived-20260413/","kind":"coverage"},{"label":"TechCrunch: OpenAI claims it solved an 80-year-old math problem, for real this time","url":"https://techcrunch.com/2026/05/20/openai-claims-it-solved-an-80-year-old-math-problem-for-real-this-time/","kind":"coverage"},{"label":"Tech Jacks: AI math reasoning milestones","url":"https://techjacksolutions.com/ai-brief/ai-math-reasoning-milestones-30-days-research-automation/","kind":"coverage"}],"url":"https://whataifound.org/finding/2026-05-erdos-unit-distance"},{"id":"2026-05-18-pevac-ps-sarbecovirus","title":"Phase 1 trial of a computationally designed pan-sarbecovirus vaccine","claim":"A vaccine whose antigen was designed entirely in silico from sarbecovirus sequence data was safe in 39 healthy volunteers and raised immune responses against SARS-CoV-2, SARS, and bat viruses that have never infected humans.","field":"medicine","date":"2026-05-18","added":"2026-08-06","lab":"University of Cambridge / DIOSynVax","model":"Unknown","verification":"peer-reviewed","autonomy":"ai-assisted","tags":["vaccine-design","clinical-trial","coronavirus","antigen-design","phase-1"],"humans":["Alasdair P. S. Munro","Jonathan L. Heeney","Saul N. Faust"],"sources":[{"label":"Journal of Infection: A phase I, needle free, dose escalation clinical trial of pEVAC-PS, a candidate pan-Sarbecovirus vaccine","url":"https://www.journalofinfection.com/article/S0163-4453(26)00084-8/fulltext","kind":"research"},{"label":"University of Cambridge: New universal vaccine technology could protect us from future virus outbreaks","url":"https://www.cam.ac.uk/research/news/new-universal-vaccine-technology-could-protect-us-from-future-virus-outbreaks","kind":"announcement"},{"label":"ScienceDaily: AI-designed universal coronavirus vaccine passes first human trial","url":"https://www.sciencedaily.com/releases/2026/06/260605023357.htm","kind":"coverage"}],"url":"https://whataifound.org/finding/2026-05-18-pevac-ps-sarbecovirus"},{"id":"2026-01-alphaevolve-bruhat","title":"Structure in Bruhat intervals of permutation groups","claim":"AlphaEvolve identified unexpected special structure in Bruhat intervals for particular permutation groups.","field":"mathematics","date":"2026-01-15","added":"2026-07-20","lab":"Google DeepMind","model":"AlphaEvolve (Gemini-based)","verification":"independent","autonomy":"search-scaffold","tags":["combinatorics","alphaevolve"],"sources":[{"label":"Ellenberg, Libedinsky, Plaza, Simental, Williamson: Bruhat intervals that are large hypercubes","url":"https://arxiv.org/abs/2601.01235","kind":"research"},{"label":"Quanta: The AI Revolution in Math Has Arrived","url":"https://www.quantamagazine.org/the-ai-revolution-in-math-has-arrived-20260413/","kind":"coverage"}],"url":"https://whataifound.org/finding/2026-01-alphaevolve-bruhat"},{"id":"2026-01-erdos-728","title":"Erdős problem #728 resolved and formalized in Lean","claim":"GPT-5.2 produced a proof of a previously open Erdős problem; Harmonic's Aristotle formalized it in Lean, making it the first Erdős problem regarded as fully resolved by AI with machine-checked proof.","field":"mathematics","date":"2026-01-13","added":"2026-07-20","lab":"OpenAI / Harmonic","model":"GPT-5.2 Pro + Aristotle","verification":"formal","autonomy":"autonomous","tags":["number-theory","erdos","lean","formalization"],"sources":[{"label":"arXiv 2601.07421: Resolution of Erdős Problem #728","url":"https://arxiv.org/abs/2601.07421","kind":"research"},{"label":"Tao: AI contributions to Erdős problems","url":"https://github.com/teorth/erdosproblems/wiki/AI-contributions-to-Erd%C5%91s-problems","kind":"research"},{"label":"The Decoder: Tao on GPT-5.2 Pro","url":"https://the-decoder.com/terence-tao-says-gpt-5-2-pro-cracked-an-erdos-problem-but-warns-the-win-says-more-about-speed-than-difficulty/","kind":"coverage"}],"url":"https://whataifound.org/finding/2026-01-erdos-728"},{"id":"2025-11-gpt5-science-acceleration","title":"Early science acceleration experiments with GPT-5","claim":"A multi-domain study documenting cases where GPT-5 contributed to research progress across mathematics, physics, biology and materials science.","field":"computer-science","date":"2025-11-20","added":"2026-07-20","lab":"OpenAI","model":"GPT-5","verification":"peer-reviewed","autonomy":"ai-assisted","tags":["meta","multi-domain"],"sources":[{"label":"arXiv 2511.16072","url":"https://arxiv.org/pdf/2511.16072","kind":"research"}],"url":"https://whataifound.org/finding/2025-11-gpt5-science-acceleration"},{"id":"2025-11-03-alphaevolve-at-scale","title":"AlphaEvolve across 67 problems: 20 improvements, 8 regressions","claim":"A systematic study with Terence Tao of an evolutionary coding agent on 67 open problems in analysis, combinatorics and geometry, reporting where it beat the literature and where it did not.","field":"mathematics","date":"2025-11-03","added":"2026-07-27","lab":"Google DeepMind","model":"AlphaEvolve (Gemini-based)","verification":"independent","autonomy":"search-scaffold","tags":["combinatorics","geometry","analysis","evolutionary-search"],"humans":["Terence Tao","Javier Gómez-Serrano","Bogdan Georgiev","Adam Zsolt Wagner"],"sources":[{"label":"arXiv: Mathematical exploration and discovery at scale","url":"https://arxiv.org/abs/2511.02864","kind":"research"},{"label":"Terence Tao's write-up","url":"https://terrytao.wordpress.com/2025/11/05/mathematical-exploration-and-discovery-at-scale/","kind":"announcement"}],"url":"https://whataifound.org/finding/2025-11-03-alphaevolve-at-scale"},{"id":"2025-10-19-gpt5-erdos-retrieval","title":"GPT-5 \"solved 10 Erdős problems\": it located existing solutions","claim":"An OpenAI executive announced that GPT-5 had solved 10 previously open Erdős problems and made progress on 11 more; the problems were only 'open' in the sense that one database maintainer was unaware of the already-published solutions the model surfaced.","field":"mathematics","date":"2025-10-19","added":"2026-07-20","lab":"OpenAI","model":"GPT-5","verification":"known","autonomy":"retrieval","tags":["number-theory","erdos","already-known","cautionary","retrieval"],"sources":[{"label":"TechCrunch: OpenAI's 'embarrassing' math","url":"https://techcrunch.com/2025/10/19/openais-embarrassing-math/","kind":"challenge"},{"label":"The Decoder: A GPT-5 math breakthrough that never happened","url":"https://the-decoder.com/leading-openai-researcher-announced-a-gpt-5-math-breakthrough-that-never-happened/","kind":"challenge"}],"url":"https://whataifound.org/finding/2025-10-19-gpt5-erdos-retrieval"},{"id":"2025-09-25-qma-amplification-limits","title":"Limits to black-box amplification in QMA, with the key step written by GPT-5","claim":"A quantum complexity paper proving black-box error amplification for QMA cannot push completeness closer to certainty than doubly exponentially, in which the pivotal technical step was produced by GPT-5 in about half an hour.","field":"computer-science","date":"2025-09-25","added":"2026-07-29","lab":"UT Austin / CWI Amsterdam","model":"GPT-5 Thinking","verification":"author-verified","autonomy":"ai-assisted","tags":["quantum-complexity","proof-assistance","oracle-separation"],"humans":["Scott Aaronson","Freek Witteveen"],"sources":[{"label":"arXiv: Limits to black-box amplification in QMA","url":"https://arxiv.org/abs/2509.21131","kind":"research"},{"label":"Scott Aaronson: The QMA Singularity","url":"https://scottaaronson.blog/?p=9183","kind":"announcement"}],"url":"https://whataifound.org/finding/2025-09-25-qma-amplification-limits"},{"id":"2025-09-17-evo-phage-genomes","title":"AI-generated bacteriophage genomes that replicate and kill bacteria","claim":"Genome language models wrote complete ΦX174-like bacteriophage genomes from scratch; of 285 designs that could be built, 16 produced viable infectious phages, several with faster lysis than the natural virus.","field":"biology","date":"2025-09-17","added":"2026-07-29","lab":"Arc Institute / Stanford University","model":"Evo 1 and Evo 2","verification":"peer-reviewed","autonomy":"ai-led","tags":["genome-design","bacteriophage","generative-biology","biosecurity"],"humans":["Samuel King","Brian Hie"],"sources":[{"label":"Science: Generative design of bacteriophages with genome language models","url":"https://www.science.org/doi/10.1126/science.aec2657","kind":"research"},{"label":"bioRxiv: Generative design of novel bacteriophages with genome language models","url":"https://www.biorxiv.org/content/10.1101/2025.09.12.675911v1","kind":"research"},{"label":"Arc Institute: How we built the first AI-generated genomes","url":"https://arcinstitute.org/news/hie-king-first-synthetic-phage","kind":"announcement"},{"label":"Asimov Press: AI-designed phages","url":"https://www.asimov.press/p/ai-phages","kind":"coverage"},{"label":"Inglesby and Hanke: Guarding against the misuse of AI-designed viruses","url":"https://www.science.org/doi/10.1126/science.aej8512","kind":"commentary"}],"url":"https://whataifound.org/finding/2025-09-17-evo-phage-genomes"},{"id":"2025-09-17-unstable-singularities","title":"New families of unstable singularities in fluid equations","claim":"Physics-informed neural networks found previously unknown unstable self-similar blow-up solutions for three fluid equations, computed to near machine precision, and revealed an unexpected near-linear relation between a solution's instability order and its blow-up rate.","field":"mathematics","date":"2025-09-17","added":"2026-07-29","lab":"Google DeepMind (with Brown, NYU and Stanford)","model":"Physics-informed neural networks with Gauss–Newton optimisation","verification":"author-verified","autonomy":"search-scaffold","tags":["pde","fluid-dynamics","singularity","physics-informed-neural-networks"],"humans":["Javier Gómez-Serrano","Tristan Buckmaster","Ching-Yao Lai","Yongji Wang"],"sources":[{"label":"arXiv: Discovery of Unstable Singularities","url":"https://arxiv.org/abs/2509.14185","kind":"research"},{"label":"DeepMind: Discovering new solutions to century-old problems in fluid dynamics","url":"https://deepmind.google/blog/discovering-new-solutions-to-century-old-problems-in-fluid-dynamics/","kind":"announcement"},{"label":"Physics World: Neural networks discover unstable singularities in fluid systems","url":"https://physicsworld.com/a/neural-networks-discover-unstable-singularities-in-fluid-systems/","kind":"coverage"}],"url":"https://whataifound.org/finding/2025-09-17-unstable-singularities"},{"id":"2025-09-11-gauss-strong-pnt","title":"Strong prime number theorem formalized in Lean by an autoformalization agent","claim":"An agent completed a Lean formalization of the strong prime number theorem in about three weeks, finishing a project two expert mathematicians had left blocked after 18 months.","field":"mathematics","date":"2025-09-11","added":"2026-07-27","lab":"Math, Inc.","model":"Gauss (autoformalization agent)","verification":"author-verified","autonomy":"ai-assisted","tags":["lean","formalization","number-theory","prime-number-theorem"],"humans":["Terence Tao","Alex Kontorovich"],"sources":[{"label":"Math, Inc.: Introducing Gauss","url":"https://www.math.inc/gauss","kind":"announcement"},{"label":"strongpnt Lean repository","url":"https://github.com/math-inc/strongpnt","kind":"research"}],"url":"https://whataifound.org/finding/2025-09-11-gauss-strong-pnt"},{"id":"2025-08-gpt5-convex-bound","title":"Improved step-size bound in smooth convex optimization","claim":"GPT-5 Pro extended a guaranteed-convexity window for gradient descent from η ≤ 1/L to η ≤ 1.5/L, but the optimal 1.75/L bound had already been published months earlier.","field":"mathematics","date":"2025-08-01","added":"2026-07-20","lab":"OpenAI","model":"GPT-5 Pro","verification":"known","autonomy":"ai-assisted","tags":["optimization","already-known","cautionary"],"humans":["Sébastien Bubeck"],"sources":[{"label":"Analysis of the claim","url":"https://hackmd.io/@niloydebbarma/HkGxVPwFgx","kind":"challenge"},{"label":"What does GPT-5's 'new math' claim actually mean?","url":"https://allthings.how/what-does-gpt-5s-new-math-claim-actually-mean/","kind":"challenge"}],"url":"https://whataifound.org/finding/2025-08-gpt5-convex-bound"},{"id":"2025-07-28-aristotle-imo-lean","title":"Machine-checked Lean proofs for five of six 2025 IMO problems","claim":"Harmonic's Aristotle produced formally verified Lean 4 proofs for five of the six 2025 IMO problems, so the gold-medal-standard score rests on a compiler rather than on human graders.","field":"mathematics","date":"2025-07-28","added":"2026-07-29","lab":"Harmonic","model":"Aristotle","verification":"formal","autonomy":"ai-led","tags":["olympiad","lean","formalization","benchmark"],"sources":[{"label":"arXiv: Aristotle: IMO-level Automated Theorem Proving","url":"https://arxiv.org/abs/2510.01346","kind":"research"},{"label":"harmonic-ai/IMO2025: Lean proofs and verification artifacts","url":"https://github.com/harmonic-ai/IMO2025","kind":"research"}],"url":"https://whataifound.org/finding/2025-07-28-aristotle-imo-lean"},{"id":"2025-07-21-gemini-deepthink-imo","title":"Gold-medal standard at the 2025 International Mathematical Olympiad","claim":"An advanced version of Gemini Deep Think read the official problem statements and wrote natural-language proofs for five of the six 2025 IMO problems inside the contest time limits, scoring 35 of 42, graded and certified by the official IMO coordinators.","field":"mathematics","date":"2025-07-21","added":"2026-07-29","lab":"Google DeepMind","model":"Gemini Deep Think (advanced version)","verification":"independent","autonomy":"ai-led","tags":["olympiad","benchmark","natural-language-proof"],"sources":[{"label":"Gemini Deep Think: official IMO 2025 solutions (PDF)","url":"https://storage.googleapis.com/deepmind-media/gemini/IMO_2025.pdf","kind":"research"},{"label":"DeepMind: Gemini with Deep Think officially achieves gold-medal standard at the IMO","url":"https://deepmind.google/discover/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/","kind":"announcement"},{"label":"Simon Willison: OpenAI's gold medal performance on the IMO","url":"https://simonwillison.net/2025/Jul/19/openai-gold-medal-math-olympiad/","kind":"commentary"}],"url":"https://whataifound.org/finding/2025-07-21-gemini-deepthink-imo"},{"id":"2025-06-03-rentosertib-phase2a","title":"Phase 2a results for a drug whose target and molecule both came from AI","claim":"Rentosertib, a TNIK inhibitor whose target was picked by one AI system and whose molecule was generated by another, improved forced vital capacity by 98.4 mL at the top dose against a 20.3 mL decline on placebo in a 71-patient randomised trial.","field":"medicine","date":"2025-06-03","added":"2026-07-29","lab":"Insilico Medicine","model":"PandaOmics (target) + Chemistry42 (molecule)","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["drug-discovery","clinical-trial","target-discovery","fibrosis"],"sources":[{"label":"Nature Medicine: A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis","url":"https://www.nature.com/articles/s41591-025-03743-2","kind":"research"},{"label":"Insilico Medicine: positive topline results of ISM001-055","url":"https://insilico.com/news/tnik-ipf-phase2a","kind":"announcement"}],"url":"https://whataifound.org/finding/2025-06-03-rentosertib-phase2a"},{"id":"2025-05-alphaevolve-kissing","title":"Improved lower bound for the 11-dimensional kissing number","claim":"AlphaEvolve improved the best known configuration for the kissing number problem in 11 dimensions.","field":"mathematics","date":"2025-05-14","added":"2026-07-20","lab":"Google DeepMind","model":"AlphaEvolve (Gemini-based)","verification":"independent","autonomy":"search-scaffold","tags":["geometry","sphere-packing","alphaevolve"],"year_posed":1694,"sources":[{"label":"AlphaEvolve paper (PDF)","url":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/AlphaEvolve.pdf","kind":"research"},{"label":"DeepMind: AlphaEvolve","url":"https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","kind":"announcement"},{"label":"IEEE Spectrum: AlphaEvolve Tackles Kissing Problem","url":"https://spectrum.ieee.org/deepmind-alphaevolve","kind":"coverage"}],"url":"https://whataifound.org/finding/2025-05-alphaevolve-kissing"},{"id":"2025-05-alphaevolve-matmul","title":"4×4 complex matrix multiplication in 48 scalar multiplications","claim":"AlphaEvolve found a scheme multiplying 4×4 complex matrices with 48 scalar multiplications, improving on Strassen's 49 from 1969.","field":"computer-science","date":"2025-05-14","added":"2026-07-20","lab":"Google DeepMind","model":"AlphaEvolve (Gemini-based)","verification":"independent","autonomy":"search-scaffold","tags":["algorithms","linear-algebra","alphaevolve"],"year_posed":1969,"sources":[{"label":"AlphaEvolve paper (PDF)","url":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/AlphaEvolve.pdf","kind":"research"},{"label":"DeepMind: AlphaEvolve","url":"https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","kind":"announcement"},{"label":"IEEE Spectrum","url":"https://spectrum.ieee.org/deepmind-alphaevolve","kind":"coverage"},{"label":"Wikipedia: AlphaEvolve","url":"https://en.wikipedia.org/wiki/AlphaEvolve","kind":"commentary"}],"url":"https://whataifound.org/finding/2025-05-alphaevolve-matmul"},{"id":"2025-05-alphaevolve-minimum-overlap","title":"Improved bound for the Erdős minimum-overlap problem","claim":"AlphaEvolve nudged the best known bound for Erdős's minimum-overlap constant, the first improvement since 2016, and sharpened several autocorrelation inequalities.","field":"mathematics","date":"2025-05-14","added":"2026-07-21","lab":"Google DeepMind","model":"AlphaEvolve (Gemini-based)","verification":"independent","autonomy":"search-scaffold","tags":["analysis","erdos","alphaevolve"],"year_posed":1955,"sources":[{"label":"DeepMind: AlphaEvolve","url":"https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","kind":"announcement"},{"label":"AlphaEvolve paper (PDF)","url":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/AlphaEvolve.pdf","kind":"research"},{"label":"Wikipedia: Minimum overlap problem","url":"https://en.wikipedia.org/wiki/Minimum_overlap_problem","kind":"commentary"}],"url":"https://whataifound.org/finding/2025-05-alphaevolve-minimum-overlap"},{"id":"2025-robin-macular","title":"Candidate treatment for dry age-related macular degeneration","claim":"The Robin system automated hypothesis generation, experiment design and data analysis, identifying a novel candidate treatment for dry AMD.","field":"biology","date":"2025-05-01","added":"2026-07-20","lab":"FutureHouse","model":"Robin (multi-agent)","verification":"peer-reviewed","autonomy":"ai-led","tags":["drug-discovery","ophthalmology","multi-agent"],"sources":[{"label":"Nature: A multi-agent system for automating scientific discovery","url":"https://www.nature.com/articles/s41586-026-10652-y","kind":"research"},{"label":"Turing Post: 12 AI Co-Scientists of 2026","url":"https://www.turingpost.com/p/ai-co-scientists-in-2026","kind":"coverage"}],"url":"https://whataifound.org/finding/2025-robin-macular"},{"id":"2025-03-12-ai-scientist-workshop-paper","title":"A fully machine-generated paper passed workshop peer review","claim":"One of three end-to-end AI-generated manuscripts submitted to an ICLR 2025 workshop averaged 6.33 from reviewers and would have been accepted; the authors withdrew it, and the bar cleared was a workshop track, not the main conference.","field":"computer-science","date":"2025-03-12","added":"2026-07-29","lab":"Sakana AI","model":"The AI Scientist-v2","verification":"author-verified","autonomy":"ai-led","tags":["autonomous-research","peer-review","agents","contested"],"sources":[{"label":"Sakana AI: The AI Scientist generates its first peer-reviewed publication","url":"https://sakana.ai/ai-scientist-first-publication/","kind":"announcement"},{"label":"TechCrunch: Sakana claims its AI paper passed peer review (it's more nuanced than that)","url":"https://techcrunch.com/2025/03/12/sakana-claims-its-ai-paper-passed-peer-review-but-its-a-bit-more-nuanced-than-that/","kind":"challenge"},{"label":"SakanaAI/AI-Scientist-ICLR2025-Workshop-Experiment (manuscripts and reviews)","url":"https://github.com/SakanaAI/AI-Scientist-ICLR2025-Workshop-Experiment","kind":"research"}],"url":"https://whataifound.org/finding/2025-03-12-ai-scientist-workshop-paper"},{"id":"2025-02-ai-coscientist-amr","title":"AI co-scientist hypotheses on antimicrobial resistance and liver fibrosis","claim":"A multi-agent system generated hypotheses that were subsequently validated experimentally in the lab.","field":"biology","date":"2025-02-19","added":"2026-07-20","lab":"Google","model":"AI Co-Scientist (Gemini 2.0 multi-agent)","verification":"peer-reviewed","autonomy":"ai-assisted","tags":["hypothesis-generation","microbiology","multi-agent"],"sources":[{"label":"Nature: Accelerating scientific discovery with Co-Scientist","url":"https://www.nature.com/articles/s41586-026-10644-y","kind":"research"},{"label":"bioRxiv: AI mirrors experimental science to uncover a novel mechanism of gene transfer","url":"https://www.biorxiv.org/content/10.1101/2025.02.19.639094v1","kind":"research"},{"label":"Turing Post: 12 AI Co-Scientists of 2026","url":"https://www.turingpost.com/p/ai-co-scientists-in-2026","kind":"coverage"}],"url":"https://whataifound.org/finding/2025-02-ai-coscientist-amr"},{"id":"2025-02-13-designed-serine-hydrolases","title":"Enzymes with working catalytic machinery designed from scratch","claim":"Diffusion-based protein design produced serine hydrolases with catalytic efficiencies up to 2.2 × 10⁵ M⁻¹s⁻¹ on five folds unlike any natural serine hydrolase, with crystal structures matching the design models to under 1 Å.","field":"chemistry","date":"2025-02-13","added":"2026-07-29","lab":"Institute for Protein Design, University of Washington","model":"RFdiffusion with PLACER ensemble scoring","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["protein-design","enzyme-design","rfdiffusion","catalysis"],"humans":["Anna Lauko","David Baker"],"sources":[{"label":"Science: Computational design of serine hydrolases","url":"https://www.science.org/doi/10.1126/science.adu2454","kind":"research"},{"label":"Baker Lab: Generating new enzymes with complex active sites","url":"https://www.bakerlab.org/2025/02/13/ai-enzymes-with-complex-active-sites/","kind":"announcement"}],"url":"https://whataifound.org/finding/2025-02-13-designed-serine-hydrolases"},{"id":"2025-01-16-esmgfp","title":"A working fluorescent protein generated by a language model","claim":"ESM3 designed esmGFP, a bright green fluorescent protein sharing only 58% of its sequence with the closest known natural fluorescent protein, a distance the authors estimate at over 500 million years of natural evolution.","field":"biology","date":"2025-01-16","added":"2026-07-29","lab":"EvolutionaryScale","model":"ESM3 (98B)","verification":"peer-reviewed","autonomy":"ai-led","tags":["protein-design","generative-biology","fluorescent-protein"],"humans":["Alexander Rives","Tom Sercu"],"sources":[{"label":"Science: Simulating 500 million years of evolution with a language model","url":"https://www.science.org/doi/10.1126/science.ads0018","kind":"research"},{"label":"EvolutionaryScale: ESM3 release","url":"https://www.evolutionaryscale.ai/blog/esm3-release","kind":"announcement"}],"url":"https://whataifound.org/finding/2025-01-16-esmgfp"},{"id":"2025-01-16-mattergen","title":"Generative model designs crystals to order; its flagship synthesis turned out to be a known compound","claim":"MatterGen generates crystal structures conditioned on target properties, and its headline experimental validation was later shown to match a compound reported in 1972 that sits in the model's own training data.","field":"materials","date":"2025-01-16","added":"2026-07-29","lab":"Microsoft Research","model":"MatterGen","verification":"disputed","autonomy":"search-scaffold","tags":["materials","crystal-structure","diffusion-model","contested"],"sources":[{"label":"Nature: A generative model for inorganic materials design","url":"https://www.nature.com/articles/s41586-025-08628-5","kind":"research"},{"label":"MatterGen code and model weights","url":"https://github.com/microsoft/mattergen","kind":"research"},{"label":"Microsoft Research: MatterGen, a new paradigm of materials design with generative AI","url":"https://www.microsoft.com/en-us/research/blog/mattergen-a-new-paradigm-of-materials-design-with-generative-ai/","kind":"announcement"},{"label":"Materials Horizons: MatterGen predicts compounds from the training dataset","url":"https://pubs.rsc.org/en/content/articlelanding/2026/mh/d6mh00268d","kind":"challenge"}],"url":"https://whataifound.org/finding/2025-01-16-mattergen"},{"id":"2024-11-20-alphaqubit","title":"A neural decoder that identifies quantum errors more accurately than hand-designed methods","claim":"A transformer trained on Google's Sycamore surface-code data cut decoding errors by 6% against tensor-network decoding and 30% against correlated matching, the best reported accuracy at the time.","field":"physics","date":"2024-11-20","added":"2026-07-29","lab":"Google DeepMind / Google Quantum AI","model":"AlphaQubit","verification":"peer-reviewed","autonomy":"ai-led","tags":["quantum-computing","error-correction","transformer"],"sources":[{"label":"Nature: Learning high-accuracy error decoding for quantum processors","url":"https://www.nature.com/articles/s41586-024-08148-8","kind":"research"},{"label":"Google: AlphaQubit, research on quantum error correction","url":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphaqubit-quantum-error-correction/","kind":"announcement"}],"url":"https://whataifound.org/finding/2024-11-20-alphaqubit"},{"id":"2024-10-02-flywire-connectome","title":"Complete wiring diagram of an adult fruit-fly brain","claim":"Machine segmentation of an electron-microscopy volume, corrected by a large community of proofreaders, produced the first synapse-level map of an entire adult brain: 139,255 neurons and roughly 50 million synapses.","field":"neuroscience","date":"2024-10-02","added":"2026-07-29","lab":"FlyWire Consortium (Princeton, MRC LMB, Cambridge, Vermont)","model":"Automated electron-microscopy segmentation networks","verification":"peer-reviewed","autonomy":"ai-assisted","tags":["connectome","electron-microscopy","segmentation","drosophila"],"humans":["Sebastian Seung","Mala Murthy","Gregory Jefferis"],"sources":[{"label":"Nature: Neuronal wiring diagram of an adult brain","url":"https://www.nature.com/articles/s41586-024-07558-y","kind":"research"},{"label":"NIH: Complete wiring map of an adult fruit fly brain","url":"https://www.nih.gov/news-events/nih-research-matters/complete-wiring-map-adult-fruit-fly-brain","kind":"announcement"},{"label":"FlyWire (data and explorer)","url":"https://flywire.ai/","kind":"research"}],"url":"https://whataifound.org/finding/2024-10-02-flywire-connectome"},{"id":"2024-07-25-alphaproof-imo","title":"Silver-medal standard at the 2024 International Mathematical Olympiad","claim":"AlphaProof and AlphaGeometry 2 together solved four of the six 2024 IMO problems for 28 of 42 points (one short of the gold threshold), with the algebra and number-theory solutions produced and checked in the Lean proof assistant.","field":"mathematics","date":"2024-07-25","added":"2026-07-20","lab":"Google DeepMind","model":"AlphaProof + AlphaGeometry 2","verification":"independent","autonomy":"ai-led","tags":["olympiad","lean","formalization","benchmark"],"sources":[{"label":"DeepMind: AI achieves silver-medal standard solving IMO problems","url":"https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/","kind":"announcement"},{"label":"Nature: Olympiad-level formal mathematical reasoning with reinforcement learning","url":"https://www.nature.com/articles/s41586-025-09833-y","kind":"research"},{"label":"Unite.AI: How AlphaProof and AlphaGeometry 2 achieved silver-medal standard","url":"https://www.unite.ai/ai-at-the-international-mathematical-olympiad-how-alphaproof-and-alphageometry-2-achieved-silver-medal-standard/","kind":"coverage"}],"url":"https://whataifound.org/finding/2024-07-25-alphaproof-imo"},{"id":"2024-05-08-alphafold3","title":"Joint structure prediction for proteins, nucleic acids and ligands (AlphaFold 3)","claim":"A diffusion-based successor to AlphaFold predicts complexes spanning proteins, DNA, RNA, small molecules and ions in a single model, with large accuracy gains over the specialised docking tools used for each case.","field":"biology","date":"2024-05-08","added":"2026-07-29","lab":"Google DeepMind / Isomorphic Labs","model":"AlphaFold 3","verification":"peer-reviewed","autonomy":"ai-led","tags":["protein-structure","drug-discovery","alphafold","diffusion-model"],"humans":["John Jumper","Max Jaderberg"],"sources":[{"label":"Nature: Accurate structure prediction of biomolecular interactions with AlphaFold 3","url":"https://www.nature.com/articles/s41586-024-07487-w","kind":"research"},{"label":"AlphaFold 3 code (google-deepmind/alphafold3)","url":"https://github.com/google-deepmind/alphafold3","kind":"research"},{"label":"Google DeepMind and Isomorphic Labs introduce AlphaFold 3","url":"https://blog.google/innovation-and-ai/products/google-deepmind-isomorphic-alphafold-3-ai-model/","kind":"announcement"}],"url":"https://whataifound.org/finding/2024-05-08-alphafold3"},{"id":"2024-02-21-tearing-mode-avoidance","title":"Reinforcement learning steers a tokamak away from tearing instabilities","claim":"A controller trained on past DIII-D shots forecast tearing-mode instabilities up to 300 ms ahead and adjusted the plasma in real time to avoid them, holding high-performance conditions that would otherwise have collapsed.","field":"physics","date":"2024-02-21","added":"2026-07-29","lab":"Princeton University / PPPL / DIII-D National Fusion Facility","model":"Deep reinforcement learning controller over a learned plasma model","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["fusion","plasma-control","reinforcement-learning","tokamak"],"humans":["Jaemin Seo","Egemen Kolemen"],"sources":[{"label":"Nature: Avoiding fusion plasma tearing instability with deep reinforcement learning","url":"https://www.nature.com/articles/s41586-024-07024-9","kind":"research"},{"label":"Princeton Engineering: Engineers use AI to wrangle fusion power for the grid","url":"https://engineering.princeton.edu/news/2024/02/21/engineers-use-ai-wrangle-fusion-power-grid","kind":"announcement"}],"url":"https://whataifound.org/finding/2024-02-21-tearing-mode-avoidance"},{"id":"2024-01-17-alphageometry","title":"Olympiad geometry solved without human demonstrations","claim":"A language model trained purely on synthetic data, paired with a symbolic deduction engine, solved 25 of 30 olympiad geometry problems against 10 for the previous best automated system and 25.9 for the average human gold medallist.","field":"mathematics","date":"2024-01-17","added":"2026-07-29","lab":"Google DeepMind","model":"AlphaGeometry","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["olympiad","geometry","neuro-symbolic","benchmark"],"humans":["Trieu Trinh","Thang Luong"],"sources":[{"label":"Nature: Solving olympiad geometry without human demonstrations","url":"https://www.nature.com/articles/s41586-023-06747-5","kind":"research"},{"label":"DeepMind: AlphaGeometry, an olympiad-level AI system for geometry","url":"https://deepmind.google/discover/blog/alphageometry-an-olympiad-level-ai-system-for-geometry/","kind":"announcement"},{"label":"google-deepmind/alphageometry (code)","url":"https://github.com/google-deepmind/alphageometry","kind":"research"}],"url":"https://whataifound.org/finding/2024-01-17-alphageometry"},{"id":"2023-12-20-antibiotic-structural-class","title":"A new structural class of antibiotic candidates against MRSA","claim":"Graph neural networks trained on 39,312 assayed compounds and applied to 12 million molecules surfaced a chemical class active against MRSA in mice, with the substructures driving each prediction made explicit rather than left opaque.","field":"chemistry","date":"2023-12-20","added":"2026-07-29","lab":"MIT / Broad Institute / Harvard","model":"Ensembles of graph neural networks with substructure attribution","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["antibiotics","graph-neural-network","explainability","mrsa"],"humans":["Felix Wong","Erica Zheng","James Collins"],"sources":[{"label":"Nature: Discovery of a structural class of antibiotics with explainable deep learning","url":"https://www.nature.com/articles/s41586-023-06887-8","kind":"research"},{"label":"ScienceDaily: Using AI, researchers identify a new class of antibiotic candidates","url":"https://www.sciencedaily.com/releases/2023/12/231221012744.htm","kind":"coverage"}],"url":"https://whataifound.org/finding/2023-12-20-antibiotic-structural-class"},{"id":"2023-12-20-coscientist","title":"A language-model agent that planned and ran chemistry experiments on lab robots","claim":"Coscientist, driven by GPT-4 with web search, documentation search, code execution and lab automation, designed and executed palladium-catalysed cross-coupling reactions on real robotic hardware from a one-line goal.","field":"chemistry","date":"2023-12-20","added":"2026-07-29","lab":"Carnegie Mellon University","model":"GPT-4 (Coscientist agent)","verification":"peer-reviewed","autonomy":"ai-led","tags":["lab-automation","agents","cross-coupling","self-driving-lab"],"humans":["Daniil Boiko","Robert MacKnight","Gabe Gomes"],"sources":[{"label":"Nature: Autonomous chemical research with large language models","url":"https://www.nature.com/articles/s41586-023-06792-0","kind":"research"},{"label":"PubMed Central full text","url":"https://pmc.ncbi.nlm.nih.gov/articles/PMC10733136/","kind":"research"},{"label":"Coscientist data and reference implementation","url":"https://github.com/gomesgroup/coscientist","kind":"research"}],"url":"https://whataifound.org/finding/2023-12-20-coscientist"},{"id":"2023-12-funsearch-binpacking","title":"Improved heuristics for online bin packing","claim":"FunSearch produced bin-packing heuristics outperforming standard baselines on benchmark distributions.","field":"computer-science","date":"2023-12-14","added":"2026-07-20","lab":"Google DeepMind","model":"FunSearch (PaLM 2 / Codey)","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["algorithms","heuristics","funsearch"],"year_posed":1971,"sources":[{"label":"Nature: Mathematical discoveries from program search with LLMs","url":"https://www.nature.com/articles/s41586-023-06924-6","kind":"research"},{"label":"FunSearch code and discovered programs","url":"https://github.com/google-deepmind/funsearch","kind":"research"}],"url":"https://whataifound.org/finding/2023-12-funsearch-binpacking"},{"id":"2023-12-funsearch-capset","title":"New lower bound constructions for the cap set problem","claim":"FunSearch discovered larger cap sets than any previously known construction, the first time an LLM-based system produced a genuinely new discovery on an established open problem.","field":"mathematics","date":"2023-12-14","added":"2026-07-20","lab":"Google DeepMind","model":"FunSearch (PaLM 2 / Codey)","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["combinatorics","cap-set","funsearch","historic-first"],"year_posed":1970,"sources":[{"label":"Nature: Mathematical discoveries from program search with LLMs","url":"https://www.nature.com/articles/s41586-023-06924-6","kind":"research"},{"label":"FunSearch code and discovered programs","url":"https://github.com/google-deepmind/funsearch","kind":"research"},{"label":"DeepMind: FunSearch","url":"https://deepmind.google/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/","kind":"announcement"}],"url":"https://whataifound.org/finding/2023-12-funsearch-capset"},{"id":"2023-11-29-a-lab-synthesis","title":"Autonomous laboratory reports solid-state synthesis of new inorganic compounds","claim":"Berkeley's A-Lab, an AI-driven autonomous laboratory, reported synthesising 41 novel inorganic compounds out of 58 targets over 17 days with minimal human intervention.","field":"materials","date":"2023-11-29","added":"2026-07-20","lab":"Lawrence Berkeley National Laboratory","model":"A-Lab (ML planning + robotics)","verification":"disputed","autonomy":"ai-led","tags":["materials","autonomous-lab","synthesis","contested"],"sources":[{"label":"Nature: An autonomous laboratory for the accelerated synthesis of inorganic materials","url":"https://www.nature.com/articles/s41586-023-06734-w","kind":"research"},{"label":"The Register: Boffins deem DeepMind's material discoveries shallow","url":"https://www.theregister.com/2024/04/11/google_deepmind_material_study/","kind":"challenge"}],"url":"https://whataifound.org/finding/2023-11-29-a-lab-synthesis"},{"id":"2023-11-29-gnome","title":"Large-scale prediction of new stable crystalline materials (GNoME)","claim":"GNoME predicted about 2.2 million new inorganic crystal structures, roughly 380,000 of them flagged as thermodynamically stable, presented as an order-of-magnitude expansion of known stable materials.","field":"materials","date":"2023-11-29","added":"2026-07-20","lab":"Google DeepMind","model":"GNoME (graph neural network)","verification":"disputed","autonomy":"search-scaffold","tags":["materials","crystal-structure","graph-neural-network","contested"],"sources":[{"label":"Nature: Scaling deep learning for materials discovery","url":"https://www.nature.com/articles/s41586-023-06735-9","kind":"research"},{"label":"GNoME models, DFT data and 380k predicted structures","url":"https://github.com/google-deepmind/materials_discovery","kind":"research"},{"label":"DeepMind: Millions of new materials discovered with deep learning","url":"https://deepmind.google/blog/millions-of-new-materials-discovered-with-deep-learning/","kind":"announcement"},{"label":"The Register: Boffins deem DeepMind's material discoveries shallow","url":"https://www.theregister.com/2024/04/11/google_deepmind_material_study/","kind":"challenge"}],"url":"https://whataifound.org/finding/2023-11-29-gnome"},{"id":"2023-11-14-graphcast","title":"Medium-range weather forecasts from a graph neural network beat the operational physics model","claim":"GraphCast produced more accurate 10-day forecasts than ECMWF's HRES on roughly 90% of the evaluated variable and lead-time combinations, in under a minute on a single TPU against hours on a supercomputer.","field":"climate","date":"2023-11-14","added":"2026-07-29","lab":"Google DeepMind","model":"GraphCast","verification":"peer-reviewed","autonomy":"ai-led","tags":["weather","forecasting","graph-neural-network"],"humans":["Remi Lam","Peter Battaglia"],"sources":[{"label":"Science: Learning skillful medium-range global weather forecasting","url":"https://www.science.org/doi/10.1126/science.adi2336","kind":"research"},{"label":"DeepMind: GraphCast, AI model for faster and more accurate global weather forecasting","url":"https://deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/","kind":"announcement"},{"label":"GraphCast code and weights (google-deepmind/weathernext)","url":"https://github.com/google-deepmind/weathernext","kind":"research"}],"url":"https://whataifound.org/finding/2023-11-14-graphcast"},{"id":"2023-09-19-alphamissense","title":"Pathogenicity predictions for 71 million human missense variants","claim":"An AlphaFold-derived model classified 89% of all 71 million possible human missense variants as likely benign or likely pathogenic, against the roughly 0.1% that carry a clinical classification today.","field":"medicine","date":"2023-09-19","added":"2026-07-29","lab":"Google DeepMind","model":"AlphaMissense","verification":"peer-reviewed","autonomy":"ai-led","tags":["genomics","variant-effect","clinical-genetics"],"sources":[{"label":"Science: Accurate proteome-wide missense variant effect prediction with AlphaMissense","url":"https://www.science.org/doi/10.1126/science.adg7492","kind":"research"},{"label":"DeepMind: A catalogue of genetic mutations to help pinpoint the cause of diseases","url":"https://deepmind.google/blog/a-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases/","kind":"announcement"},{"label":"Nature Biotechnology: Advancing missense variant pathogenicity prediction","url":"https://www.nature.com/articles/s41587-023-01999-y","kind":"research"},{"label":"AlphaMissense predictions and code","url":"https://github.com/google-deepmind/alphamissense","kind":"research"}],"url":"https://whataifound.org/finding/2023-09-19-alphamissense"},{"id":"2023-06-07-alphadev","title":"Faster sorting routines discovered and merged into the LLVM C++ library","claim":"AlphaDev found shorter branchless routines for sorting small fixed-length inputs; reverse-engineered to C++, they were merged into LLVM's libc++, the first change to those routines in over a decade.","field":"computer-science","date":"2023-06-07","added":"2026-07-20","lab":"Google DeepMind","model":"AlphaDev (AlphaZero-based)","verification":"independent","autonomy":"search-scaffold","tags":["algorithms","sorting","reinforcement-learning","deployed"],"sources":[{"label":"Nature: Faster sorting algorithms discovered using deep reinforcement learning","url":"https://www.nature.com/articles/s41586-023-06004-9","kind":"research"},{"label":"AlphaDev code and discovered sorting routines","url":"https://github.com/google-deepmind/alphadev","kind":"research"},{"label":"DeepMind: AlphaDev discovers faster sorting algorithms","url":"https://deepmind.google/blog/alphadev-discovers-faster-sorting-algorithms/","kind":"announcement"},{"label":"Wikipedia: AlphaDev","url":"https://en.wikipedia.org/wiki/AlphaDev","kind":"commentary"}],"url":"https://whataifound.org/finding/2023-06-07-alphadev"},{"id":"2022-10-05-alphatensor","title":"Faster matrix-multiplication algorithms found by reinforcement learning","claim":"AlphaTensor discovered matrix-multiplication schemes using fewer scalar multiplications than any previously known, including a way to multiply 4×4 matrices over GF(2) in 47 multiplications versus the 49 of two-level Strassen.","field":"computer-science","date":"2022-10-05","added":"2026-07-20","lab":"Google DeepMind","model":"AlphaTensor (AlphaZero-based)","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["algorithms","linear-algebra","reinforcement-learning"],"year_posed":1969,"sources":[{"label":"Nature: Discovering faster matrix multiplication algorithms with reinforcement learning","url":"https://www.nature.com/articles/s41586-022-05172-4","kind":"research"},{"label":"AlphaTensor code and discovered algorithms","url":"https://github.com/google-deepmind/alphatensor","kind":"research"},{"label":"DeepMind: Discovering novel algorithms with AlphaTensor","url":"https://deepmind.google/discover/blog/discovering-novel-algorithms-with-alphatensor/","kind":"announcement"},{"label":"Quanta: AI Reveals New Possibilities in Matrix Multiplication","url":"https://www.quantamagazine.org/ai-reveals-new-possibilities-in-matrix-multiplication-20221123/","kind":"coverage"}],"url":"https://whataifound.org/finding/2022-10-05-alphatensor"},{"id":"2022-02-16-tokamak-plasma-control","title":"Deep reinforcement learning controls tokamak fusion plasma","claim":"A reinforcement-learning controller learned to shape and stabilise the magnetic confinement of real fusion plasma inside the TCV tokamak, holding configurations that are hard to sustain by hand.","field":"physics","date":"2022-02-16","added":"2026-07-21","lab":"Google DeepMind","model":"DeepMind RL controller","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["physics","fusion","reinforcement-learning","control"],"sources":[{"label":"Nature: Magnetic control of tokamak plasmas through deep reinforcement learning","url":"https://www.nature.com/articles/s41586-021-04301-9","kind":"research"},{"label":"DeepMind: Accelerating fusion science through learned plasma control","url":"https://deepmind.google/discover/blog/accelerating-fusion-science-through-learned-plasma-control/","kind":"announcement"},{"label":"EPFL: EPFL and DeepMind use AI to control plasmas for fusion","url":"https://actu.epfl.ch/news/epfl-and-deepmind-use-ai-to-control-plasmas-for-nu/","kind":"announcement"}],"url":"https://whataifound.org/finding/2022-02-16-tokamak-plasma-control"},{"id":"2021-12-01-knot-theory-intuition","title":"Two theorems found by machine pattern-spotting in knot theory and representation theory","claim":"Networks trained to predict one mathematical invariant from another, then probed for which inputs mattered, pointed mathematicians to a proved theorem linking a knot's signature to a new geometric quantity, and to structure behind the combinatorial invariance conjecture.","field":"mathematics","date":"2021-12-01","added":"2026-07-29","lab":"Google DeepMind (with Oxford and Sydney)","model":"Supervised networks with gradient-based attribution","verification":"peer-reviewed","autonomy":"ai-assisted","tags":["knot-theory","representation-theory","conjecture-generation"],"humans":["Alex Davies","Marc Lackenby","Geordie Williamson"],"sources":[{"label":"Nature: Advancing mathematics by guiding human intuition with AI","url":"https://www.nature.com/articles/s41586-021-04086-x","kind":"research"},{"label":"Knot theory and representation theory notebooks","url":"https://github.com/google-deepmind/mathematics_conjectures","kind":"research"},{"label":"University of Oxford: Machine learning helps mathematicians make new connections","url":"https://www.ox.ac.uk/news/2021-12-01-machine-learning-helps-mathematicians-make-new-connections-0","kind":"announcement"},{"label":"Ernest Davis: Deep Learning and Mathematical Intuition (review)","url":"https://arxiv.org/abs/2112.04324","kind":"challenge"}],"url":"https://whataifound.org/finding/2021-12-01-knot-theory-intuition"},{"id":"2021-07-15-alphafold2","title":"Accurate protein structure prediction across the known proteome (AlphaFold2)","claim":"AlphaFold2 predicted three-dimensional structures for nearly all catalogued proteins from amino-acid sequence at accuracy rivalling experiment, work that earned the 2024 Nobel Prize in Chemistry.","field":"biology","date":"2021-07-15","added":"2026-07-21","lab":"Google DeepMind","model":"AlphaFold2","verification":"independent","autonomy":"ai-led","tags":["protein-structure","biology","alphafold","nobel-prize"],"humans":["John Jumper","Demis Hassabis"],"year_posed":1972,"sources":[{"label":"Nature: Highly accurate protein structure prediction with AlphaFold","url":"https://www.nature.com/articles/s41586-021-03819-2","kind":"research"},{"label":"AlphaFold code (google-deepmind/alphafold)","url":"https://github.com/google-deepmind/alphafold","kind":"research"},{"label":"DeepMind: AlphaFold","url":"https://deepmind.google/science/alphafold/","kind":"announcement"},{"label":"Nobel Prize: Chemistry 2024 press release","url":"https://www.nobelprize.org/prizes/chemistry/2024/press-release/","kind":"commentary"}],"url":"https://whataifound.org/finding/2021-07-15-alphafold2"},{"id":"2021-04-29-wagner-combinatorics","title":"Reinforcement learning refutes several conjectures in extremal combinatorics","claim":"A neural network building graphs edge by edge under a cross-entropy search produced explicit counterexamples to several published conjectures in graph theory and extremal combinatorics.","field":"mathematics","date":"2021-04-29","added":"2026-07-29","lab":"Tel Aviv University","model":"Deep cross-entropy method (custom network)","verification":"independent","autonomy":"search-scaffold","tags":["combinatorics","graph-theory","counterexample","reinforcement-learning"],"humans":["Adam Zsolt Wagner"],"sources":[{"label":"arXiv: Constructions in combinatorics via neural networks","url":"https://arxiv.org/abs/2104.14516","kind":"research"},{"label":"Wagner's cross-entropy code for the counterexamples","url":"https://github.com/zawagner22/cross-entropy-for-combinatorics","kind":"research"},{"label":"Refutation of Spectral Graph Theory Conjectures with Monte Carlo Search","url":"https://arxiv.org/abs/2207.03343","kind":"challenge"}],"url":"https://whataifound.org/finding/2021-04-29-wagner-combinatorics"},{"id":"2020-02-20-halicin","title":"Halicin, an antibiotic found by a neural network screening a compound library","claim":"A graph neural network trained on 2,335 molecules picked halicin out of a repurposing library; it killed multidrug-resistant bacteria including Acinetobacter baumannii and Mycobacterium tuberculosis in vitro and cleared infections in mice.","field":"medicine","date":"2020-02-20","added":"2026-07-29","lab":"MIT / Broad Institute","model":"Directed message-passing graph neural network (Chemprop)","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["antibiotics","drug-repurposing","graph-neural-network"],"humans":["Jonathan Stokes","Regina Barzilay","James Collins"],"sources":[{"label":"Cell: A deep learning approach to antibiotic discovery","url":"https://www.cell.com/cell/fulltext/S0092-8674(20)30102-1","kind":"research"},{"label":"MIT News: Artificial intelligence yields new antibiotic","url":"https://news.mit.edu/2020/artificial-intelligence-identifies-new-antibiotic-0220","kind":"announcement"}],"url":"https://whataifound.org/finding/2020-02-20-halicin"},{"id":"2017-12-14-kepler-90i","title":"An eighth planet around Kepler-90 found by a neural network","claim":"A convolutional network re-examining Kepler light curves recovered transit signals the standard pipeline had ranked below threshold, yielding Kepler-90i and making Kepler-90 the first star besides the Sun known to host eight planets.","field":"astronomy","date":"2017-12-14","added":"2026-07-29","lab":"Google Brain / University of Texas at Austin","model":"Convolutional neural network (AstroNet)","verification":"peer-reviewed","autonomy":"search-scaffold","tags":["exoplanets","kepler","transit-detection"],"humans":["Christopher Shallue","Andrew Vanderburg"],"sources":[{"label":"The Astronomical Journal: Identifying Exoplanets with Deep Learning","url":"https://iopscience.iop.org/article/10.3847/1538-3881/aa9e09","kind":"research"},{"label":"AstroNet: Kepler light-curve models (exoplanet-ml)","url":"https://github.com/google-research/exoplanet-ml","kind":"research"},{"label":"NASA: Artificial intelligence, NASA data used to discover eighth planet circling distant star","url":"https://www.nasa.gov/news-release/artificial-intelligence-nasa-data-used-to-discover-eighth-planet-circling-distant-star/","kind":"announcement"}],"url":"https://whataifound.org/finding/2017-12-14-kepler-90i"}]}