Events per month
Year
Research field
Color by
Sort by
- Landmark discovery claim
- Direct discovery
- Measured acceleration
- Capability benchmark
- Field diffusion
- Infrastructure
- Adoption context
- Limitation or failure
Swipe across the graph to explore the full timeline, or choose a year above.
Focus a dot for a preview, use ← and → to move between dots, and press Enter to open the event.
Reading the pattern
Denser clusters mean more events recorded in that period. This is a curated ledger: coverage, publication timing, and selection affect the pattern. Event counts alone do not establish scientific progress or acceleration.
For measured outcomes over time, explore the Indicators. For comparisons of AI’s contribution, see the AI effect studies.
Event list
- Claude-assisted pipeline releases a UV sky map with explicit measured and predicted layersDirect discovery
- DOE awards $159M to twelve new Phase II Genesis science projectsAdoption context
- DOE and NIH join Biohub in a national biological-data foundation with federal investmentAdoption context
- SciExam finds six agent-built ENSO models exceed a published reference on hidden testsCapability benchmark
- Gan Jiang improves frozen diffraction-analysis workflows on held-out samplesCapability benchmark
- OpenAI releases 722 mathematical manuscripts and selected proof artifacts for community assessmentLandmark discovery claim
- TasteVal isolates experiment-planning efficiency and compares research agents with human expertsCapability benchmark
- 2D-FET-Bench finds expert acceptance well below automated layout successCapability benchmark
- CrystalJev tests single-pass materials screening with prospective DFT labelsMeasured acceleration
- BootLoops reports fifteen new Feynman-integral results with standalone numerical checksDirect discovery
- AI Theorist develops an excitonic interpretation of new ruthenium-chloride spectraDirect discovery
- SynthID Bio preserves protein function while embedding detectable provenanceInfrastructure
- CompMat-Bench finds scientific errors persist despite high guided materials-task successLimitation or failure
- EurekaBench finds research agents match prediction more closely than scientific insightLimitation or failure
- R-GroundBench reveals a recognition-to-editing gap in patent molecular structuresLimitation or failure
- IsoDDE reports wet-lab validation after multi-day autonomous molecular searchMeasured acceleration
- MatToolBench exposes low agent success on professional materials softwareLimitation or failure
- Quine prioritizes compounds that produce predicted pancreatic tumor-state shiftsDirect discovery
- LLM experiment planners fail to beat a statistical baseline across biochemical search tasksLimitation or failure
- Claude autonomously computes a nine-loop amplitude in planar N=4 super-Yang-MillsDirect discovery
- AlphaFold Database adds viral protein complexes for pandemic researchInfrastructure
- ExplorationBench tests whether agents can learn unfamiliar rules through experimentsCapability benchmark
- AI-directed lab discovers durable palladium-oxide catalysts for acidic water electrolysisDirect discovery
- Claude agents identify array-associated reverse transcriptases in phage genomesDirect discovery
- Replicated materials agents recover a real frontier while converging on the same invalid structureLimitation or failure
- RetroChimera improves expert-rated retrosynthesis through complementary model ensemblingCapability benchmark
- Claude optimizes biomolecular inference models with released code and measured runtime gainsMeasured acceleration
- Google and MIT map scientific AI use through Gemini telemetry, model citations and researcher surveysField diffusion
- Paper2Agent evaluates executable paper reuse across computational research repositoriesCapability benchmark
- SynAgent revises thin-film growth hypotheses through an 18-experiment physical campaignDirect discovery
- AlphaGenome Atlas releases genome-wide variant predictions with research case studiesInfrastructure
- OpenAI releases an AI-generated forced Navier–Stokes blowup proof claimLandmark discovery claim
- OpenAI releases an AI-generated unforced Euler blowup proof claimLandmark discovery claim
- Alpöge and Buckmaster publish AI-assisted smooth-forcing Boussinesq blowup constructionDirect discovery
- OpenAI measures coding-agent adoption across its research organizationField diffusion
- Claude completes an independently checked Lean formalization of Fermat’s Last TheoremMeasured acceleration
- Astra strengthens an asymptotic lower bound on large gaps between primesDirect discovery
- AI-assisted reconstruction yields a complete male fruit-fly brain and nerve-cord connectomeInfrastructure
- MAPL-EMIT automates global methane-plume mapping from hyperspectral satellite dataInfrastructure
- Autoresearch Bench measures four-hour agent experimentation loopsCapability benchmark
- Expert-guided AutoDiscovery surfaces a lobular breast-cancer immune signal validated in tissueDirect discovery
- Astra supplies a proof claiming infinitely many prime gaps of at most 186Direct discovery
- AutoXRD benchmarks agents on executable powder-diffraction workflowsCapability benchmark
- ML research agents explore novel solutions but rarely convert novelty into performanceCapability benchmark
- Gemini Co-Scientist moves from hypotheses to execution-grounded researchMeasured acceleration
- ABE-Ralph exposes methodological hallucinations in agent-run research reproductionsLimitation or failure
- Human–AxiomProver collaboration proves the Lyons–White rate-monotonicity classificationDirect discovery
- BixBench3 scales biology-agent evaluation to complete computational studiesCapability benchmark
- Planetary Prediction Engine improves autonomous geospatial modeling benchmarksCapability benchmark
- Claude raises a Riemann-zeta critical-line bound from 41.6% to 67.2%Direct discovery
- WeatherNext Cyclones gains more than a day of forecast lead timeMeasured acceleration
- OpenAI reports Astra generated ten Lean-certified mathematical advancesLandmark discovery claim
- Shadow evaluations find agents miss the core of open-ended AI researchLimitation or failure
- Claude-assisted cryptanalysis weakens HAWK and reduced-round AESDirect discovery
- Global survey measures AI use and trust gaps in research workflowsField diffusion
- AI-guided work produces five new Banach-space theory resultsDirect discovery
- SparksMatter proposes CaMg₂Si₂ thermoelectric candidates through an autonomous in-silico loopDirect discovery
- BAISBench evaluates AI scientists on real single-cell discovery tasksCapability benchmark
- CausalGame tests whether AI scientist agents can reason causallyCapability benchmark
- PHREEQC-MCQ-200 diagnoses scientific simulator tool-use agentsCapability benchmark
- Neural iterative design creates high-affinity drug-binding proteinsDirect discovery
- PhySciBench tests deep-research agents in physics and chemistry workflowsCapability benchmark
- SciRisk-Bench maps safety failures in science-oriented LLMsLimitation or failure
- LifeSciBench evaluates agents on expert life-science research workflowsCapability benchmark
- AxiomProver formalizes components of new proofs of Thakur’s power-sum hypothesesDirect discovery
- SciAgentArena benchmarks agents on cross-scale biomedical research challengesCapability benchmark
- InquiTree finds research agents lose critical judgment in inquiry loopsLimitation or failure
- OpenAlex study links AI use to higher measured scientific creativityField diffusion
- Human–AxiomProver collaboration constructs a bijection for the Andrews–Dhar partition problemDirect discovery
- Phenotypic AI generates and tests cell-type-selective small moleculesDirect discovery
- SpatialBench-Long probes agents on end-to-end spatial biology reasoningCapability benchmark
- Large-scale study finds AI research agents narrow scientific explorationLimitation or failure
- Rethlas generates expert-checked proofs for eight open algebra problemsDirect discovery
- SMDD-Bench tests agents on multi-turn small-molecule drug designCapability benchmark
- Generative models design experimentally verified small-molecule binding proteinsDirect discovery
- An OpenAI model disproves the planar unit-distance conjectureLandmark discovery claim
- BioXArena benchmarks agents on biomedical ML pipeline tasksCapability benchmark
- NSF X-Labs funds milestone-driven research organizationsAdoption context
- Benchling Model Hub embeds scientific AI models in R&D recordsInfrastructure
- PoLARIS autonomously optimizes lead-free perovskite nanocrystals in hoursMeasured acceleration
- SpeciesNet removes a months-long camera-trap analysis bottleneckMeasured acceleration
- Large-scale audit finds hallucinated citations entering scientific literatureLimitation or failure
- OpenAlex study maps AI adoption across 227M scholarly worksField diffusion
- Lancet audit finds fabricated citations rising in biomedical papersLimitation or failure
- Nature surveys early measurements of AI-generated scientific writingField diffusion
- NBER study finds AlphaFold shifted structural-biology research directionField diffusion
- AI literature tools fail to reliably flag retracted papersLimitation or failure
- Biohub commits $500M to a Virtual Biology Initiative for AI-accelerated cell modelsAdoption context
- AutoResearchBench tests agents on scientific literature discoveryCapability benchmark
- Synthegy uses language models for strategy-aware synthesis planningCapability benchmark
- Medra Lab 001 scales an autonomous biology lab platformMeasured acceleration
- GeneBench measures multi-stage genomics and quantitative biology inferenceCapability benchmark
- GPT-Rosalind targets life-sciences research workflowsCapability benchmark
- Machine-learning-designed receptors become artificial protein switchesDirect discovery
- Amazon Bio Discovery links AI biology models to wet-lab validationDirect discovery
- Stanford AI Index tracks AI-for-science diffusion and agent limitsField diffusion
- Research Policy study maps uneven AI benefits across scientific fieldsField diffusion
- DARPA MATHBAC targets agentic AI platforms for scientific discoveryAdoption context
- NIH pilots point toward an AI assurance lab for biomedical researchLimitation or failure
- CellVoyager autonomously analyzes single-cell biology dataInfrastructure
- DOE opens $293M Genesis Mission RFA for AI-enabled science challengesAdoption context
- AI agents are tested for operating advanced scientific instrumentsCapability benchmark
- DINOv3 improves open global canopy-height mappingInfrastructure
- AI-assisted search claims analytic solutions for cosmic-string radiationDirect discovery
- Evo 2 scales genomic foundation models across domains of lifeCapability benchmark
- DataJoint adds an agentic control layer for reproducible research workflowsMeasured acceleration
- PNAS study finds journal AI policies leave a disclosure gapLimitation or failure
- Language-model meta-design generates families of quantum experimentsDirect discovery
- UKRI sets its first AI strategy for research and innovationAdoption context
- USGS publishes an agency strategy for AI-enabled scienceLimitation or failure
- Self-driving tableting data factory automates pharmaceutical formulation searchMeasured acceleration
- CESifo paper measures AlphaFold productivity gains and inequality among scientistsField diffusion
- Review sets criteria for atomistic foundation models in chemistry and materialsAdoption context
- DeepMind's Aletheia produces autonomous papers and open-problem solutionsLandmark discovery claim
- AIRS-Bench evaluates agents across the machine-learning research lifecycleCapability benchmark
- GPT-5 autonomous lab lowers cell-free protein synthesis costMeasured acceleration
- OpenScholar builds an open retrieval-augmented system for literature synthesisCapability benchmark
- LABBench2 raises the bar for biology research-agent evaluationCapability benchmark
- FIRE-Bench measures full-cycle rediscovery of scientific insightsCapability benchmark
- FrontierScience evaluates models on expert-level STEM research tasksCapability benchmark
- AlphaGenome predicts regulatory variant effects across genomic modalitiesCapability benchmark
- ACL-family conference papers show rising hallucinated citationsLimitation or failure
- Google.org names 12 AI for Science fund awardeesAdoption context
- Deep Research evaluates interactive multi-agent scientific workflowsCapability benchmark
- Nature study finds AI expands scientists' impact but narrows science's focusField diffusion
- DeepMet reveals previously unrecognized mammalian metabolitesDirect discovery
- HeurekaBench grounds co-scientist evaluation in data-backed single-cell workflowsCapability benchmark
- AlphaFold impact report links diffusion to publications, structures, and patentsField diffusion
- National Academies report recommends hybrid foundation models for DOE scienceLimitation or failure
- Self-driving lab discovers nanophotonic emission-steering rulesMeasured acceleration
- German research-organization survey finds AI use in core scientific tasksField diffusion
- White House launches the Genesis Mission for AI-accelerated scienceAdoption context
- Study measures rapid foundation-model uptake across scientific papersField diffusion
- UK publishes a national AI for Science StrategyAdoption context
- Evo genomic model produces functional de novo genesDirect discovery
- AI co-scientist hypothesis matches bacterial gene-transfer mechanismDirect discovery
- AlphaProof uses reinforcement learning for formal mathematical proofsCapability benchmark
- Kosmos runs long-horizon autonomous discovery cycles with traceable reportsInfrastructure
- Elsevier survey measures accelerating AI use and weak governance supportField diffusion
- Active-learning workflow narrows nanomedicine formulation searchMeasured acceleration
- European Commission launches an AI in Science Strategy centered on RAISEAdoption context
- Wiley survey reports broad researcher AI adoption with rising cautionField diffusion
- Robots map chemical reaction hyperspaces at scaleMeasured acceleration
- CRESt discovers an eight-element catalyst through multimodal active learning and robotic testsDirect discovery
- Genomic language models generate viable bacteriophage genomesDirect discovery
- AI Scientist systems show hidden workflow failure modesLimitation or failure
- DLESyM simulates 1,000 years of current climate in hoursMeasured acceleration
- Study finds collaboration networks shape AI adoption in scienceField diffusion
- AlphaFold collaboration study finds limited interdisciplinary spilloverField diffusion
- Physics-tailored machine learning reveals dusty-plasma force lawsDirect discovery
- OpenCRISPR-1 shows AI-generated genome editing in human cellsDirect discovery
- NSF funds a materials-focused National AI Research InstituteAdoption context
- Virtual Lab designs experimentally active SARS-CoV-2 nanobodiesDirect discovery
- Chemma adapts large language models for organic synthesis tasksCapability benchmark
- Autonomous enzyme-engineering platform improves two enzymes in four weeksMeasured acceleration
- Chemical language model yields nanomolar A2A receptor ligandsDirect discovery
- Matbench Discovery standardizes crystal-stability evaluation for materials AICapability benchmark
- AlphaEvolve generalizes coding agents for algorithm discovery and optimizationMeasured acceleration
- AI-discovered rentosertib reaches a randomized phase 2a trialDirect discovery
- Backfill study argues AI Scientists fail at implementation and verificationLimitation or failure
- ScienceBoard evaluates computer-using agents in realistic scientific workflowsCapability benchmark
- Robin links research agents to lab validation for macular-degeneration discoveryLandmark discovery claim
- OMol25 scales open molecular DFT data for chemistry modelsCapability benchmark
- NHANES study links AI-ready datasets to formulaic biomedical papersLimitation or failure
- AMIE evaluates conversational diagnostic AI with simulated patientsCapability benchmark
- PaperBench measures whether agents can replicate AI research papersCapability benchmark
- BixBench evaluates agents on open-ended computational-biology analysisCapability benchmark
- AI co-scientist frames hypothesis generation as a multi-agent research workflowCapability benchmark
- RFdiffusion-based workflow designs catalytic serine hydrolasesDirect discovery
- MatterGen uses diffusion modelling for inverse inorganic materials designDirect discovery
- Universal machine-learned potentials show systematic softeningDirect discovery
- Nucleotide Transformer benchmarks foundation models for human genomicsCapability benchmark
- RE-Bench compares AI agents with humans on AI R&D tasksCapability benchmark
- AlphaQubit learns quantum error decoding from processor dataMeasured acceleration
- TrialGPT uses language models for patient-to-trial matchingCapability benchmark
- Mobile robots perform exploratory synthetic chemistry workflowsMeasured acceleration
- OMat24 releases open inorganic-materials data and pretrained modelsCapability benchmark
- AI-assisted microfluidics screens 10,000 photocatalytic conditions per dayMeasured acceleration
- Study quantifies AI use and citation-associated benefits across scienceField diffusion
- Symbolic transformers discover new Lyapunov functions for dynamical systemsDirect discovery
- MLE-bench evaluates agents on Kaggle-style machine-learning engineeringCapability benchmark
- Chemistry Nobel recognizes computational protein design and AlphaFold structure predictionAdoption context
- ScienceAgentBench tests agents on data-driven discovery tasksCapability benchmark
- Singapore commits S$120M to a national AI for Science initiativeAdoption context
- EMA publishes a reflection paper on AI across the medicinal product lifecycleLimitation or failure
- TxGNN frames drug repurposing as zero-shot graph predictionInfrastructure
- Systematic review finds weak baselines in most claimed ML fluid-solver speedupsLimitation or failure
- NSF and Simons Foundation launch astronomy-focused AI institutesAdoption context
- DSBench tests data-science agents on more realistic tasksCapability benchmark
- Closed-loop transfer uses AI experiments to extract chemical knowledgeMeasured acceleration
- The AI Scientist prototypes an automated machine-learning research loopInfrastructure
- Few-shot AI designs antibacterial HDP-mimicking polymersMeasured acceleration
- SciCode evaluates models on scientist-curated research codingCapability benchmark
- DOE roadmap proposes FASST infrastructure for scientific AIAdoption context
- LAB-Bench evaluates language agents on practical biology research tasksCapability benchmark
- OpenAI and Los Alamos launch bioscience lab evaluations for frontier modelsAdoption context
- AI and cloud HPC narrow battery-electrolyte candidates for synthesisDirect discovery
- Knowledge-graph reasoning generates drug-discovery evidence chainsDirect discovery
- DiscoveryBench evaluates LLM agents on data-driven discovery tasksCapability benchmark
- DiscoveryWorld benchmarks end-to-end scientific discovery agentsCapability benchmark
- Royal Society report assesses opportunities and risks for AI in scienceLimitation or failure
- Prov-GigaPath scales whole-slide foundation models for pathologyCapability benchmark
- AlphaFold 3 extends structure prediction to biomolecular complexesInfrastructure
- ChemCrow equips language models with chemistry toolsInfrastructure
- SWE-agent tests agent-computer interfaces for repository-level codingCapability benchmark
- POLYGON generates and tests dual-target drug-like compoundsDirect discovery
- data-to-paper links LLM research automation to traceable manuscriptsInfrastructure
- EU scientific advisers recommend infrastructure for AI uptake in scienceLimitation or failure
- Stanford AI Index adds a science and medicine chapterLimitation or failure
- Nature Reviews Physics surveys neural operators for scientific simulationField diffusion
- Perspective argues for rigor-first AI in theoretical physics and mathematicsLimitation or failure
- RIKEN launches an AI for Science Platform DivisionAdoption context
- Active learning guides robotic discovery of redox-flow electrolyte solventsMeasured acceleration
- SyntheMol designs synthesizable antibiotic candidatesDirect discovery
- European Research Area issues guidelines for generative AI in researchAdoption context
- UNI scales self-supervised learning for computational pathologyCapability benchmark
- Robotics and machine learning discover all-natural plastic substitutesDirect discovery
- RoseTTAFold All-Atom models biomolecular assemblies beyond proteinsInfrastructure
- Data Interpreter builds an LLM agent for end-to-end data scienceCapability benchmark
- ModelAngelo automates atomic model building in cryo-EM mapsCapability benchmark
- scGPT adapts generative pretrained transformers to single-cell biologyCapability benchmark
- ChatGPT Advanced Data Analysis is tested on clinical-study machine learningMeasured acceleration
- Deep reinforcement learning helps avoid tearing instabilities in fusion plasmaMeasured acceleration
- Self-driving-lab perspective proposes metrics for chemistry and materials automationInfrastructure
- Programmable reaction engine integrates synthesis and self-optimizationMeasured acceleration
- UKRI funds AI research hubs for science, engineering and real-world dataAdoption context
- NSF launches NAIRR pilot for shared AI research infrastructureAdoption context
- Dynamic knowledge graphs coordinate distributed self-driving labsMeasured acceleration
- MedSAM adapts segment-anything models to medical imagesCapability benchmark
- WHO issues guidance on large multimodal models in healthLimitation or failure
- AlphaGeometry pairs language models with symbolic deduction for geometry proofsCapability benchmark
- InfiAgent-DABench evaluates agents on executable data-analysis tasksCapability benchmark
- Isomorphic Labs signs AI drug-discovery collaborations with Lilly and NovartisAdoption context
- Coscientist links large language models to automated chemistry experimentsInfrastructure
- FunSearch uses language models and program search for mathematical discoveryDirect discovery
- NSF restricts generative AI use in merit reviewAdoption context
- Deep learning designs cell-type-specific synthetic enhancersMeasured acceleration
- PaperQA applies retrieval-augmented generation to scientific literature questionsCapability benchmark
- A-Lab demonstrates autonomous inorganic-materials synthesisMeasured acceleration
- GNoME scales AI-guided discovery of stable inorganic crystalsDirect discovery
- GPQA benchmarks models on expert-written graduate science questionsCapability benchmark
- GraphCast shows AI competing with operational weather forecastingMeasured acceleration
- Argonne announces the Trillion Parameter Consortium for scientific AI modelsAdoption context
- Deep learning and CFPS compress antimicrobial-peptide screeningMeasured acceleration
- OpenDAC releases a large dataset for direct-air-capture sorbentsCapability benchmark
- UKRI launches the first phase of the UK AI Research ResourceAdoption context
- EVEscape forecasts viral escape from prepandemic sequence dataInfrastructure
- SWE-bench turns real GitHub issues into a coding-agent benchmarkCapability benchmark
- MLAgentBench evaluates agents on machine-learning experimentationCapability benchmark
- Nature survey captures early post-ChatGPT researcher expectations for AI in scienceField diffusion
- AdsorbML accelerates adsorption-energy calculations for catalysisCapability benchmark
- AlphaMissense releases proteome-wide missense variant predictionsInfrastructure
- Microsoft launches DeepSpeed4Science for AI-enabled scientific modelsAdoption context
- ChatGPT struggles to attach real evidence to domain answersDirect discovery
- CHGNet adds charge-informed dynamics to universal materials potentialsInfrastructure
- RETFound trains a retinal-image foundation model for disease detectionCapability benchmark
- Patterns paper frames data leakage as an ML-based science reproducibility crisisLimitation or failure
- UNESCO publishes global guidance on generative AI in education and researchLimitation or failure
- BioCoder evaluates language models on bioinformatics code generationCapability benchmark
- Med-PaLM M prototypes a multimodal generalist biomedical modelCapability benchmark
- Med-PaLM tests large language models on medical knowledge tasksCapability benchmark
- RFdiffusion applies diffusion models to de novo protein designDirect discovery
- LeanDojo opens infrastructure for retrieval-augmented theorem provingCapability benchmark
- OECD report frames AI in science as a policy and infrastructure challengeLimitation or failure
- REMEDIS improves data-efficient diagnostic imaging modelsCapability benchmark
- AlphaDev discovers faster low-level sorting routinesDirect discovery
- Geneformer adapts foundation-model transfer learning to network biologyCapability benchmark
- Reference hallucination becomes a measurable LLM failure modeLimitation or failure
- Machine learning identifies abaucin as a targeted antibiotic leadDirect discovery
- Active learning compresses boundary-layer wind-tunnel explorationMeasured acceleration
- ChatGPT-generated medical content shows high reference error ratesDirect discovery
- HaluEval benchmarks hallucination detection in large language modelsCapability benchmark
- FDA opens policy discussion on AI in drug development and manufacturingLimitation or failure
- NSF expands the National AI Research Institutes networkAdoption context
- Protein language models guide antibody affinity maturationMeasured acceleration
- BacterAI accelerates microbial metabolism mapping through active learningMeasured acceleration
- Active learning steers neutron spectroscopy measurementsMeasured acceleration
- AI-Descartes combines data and axioms for scientific law discoveryCapability benchmark
- Semantic Reader explores AI-augmented interfaces for reading papersDirect discovery
- ESMFold predicts metagenomic protein structures from language-model representationsInfrastructure
- LIDA uses language models to generate data visualizationsCapability benchmark
- Deep learning supports de novo design of artificial luciferasesMeasured acceleration
- Deep learning filters technosignature candidates in SETI dataMeasured acceleration
- U.S. and EU sign an AI public-good research arrangementAdoption context
- ProGen generates functional protein sequences across familiesDirect discovery
- Nature and Springer Nature set early LLM authorship and disclosure rulesAdoption context
- Transition1x targets reactive chemistry for machine-learned potentialsInfrastructure
- M3GNet scales graph neural interatomic potentials across the periodic tableInfrastructure
- DS-1000 benchmarks language models on data-science codeCapability benchmark
- Galactica trains a language model on scientific corporaCapability benchmark
- Schmidt Futures funds a global AI in Science postdoctoral fellowshipAdoption context
- Draft, Sketch, and Prove uses informal proofs to guide formal proversCapability benchmark
- AlphaTensor uses reinforcement learning to discover matrix multiplication algorithmsDirect discovery
- DiffDock reframes molecular docking as a diffusion problemCapability benchmark
- Robotic experimentation optimizes non-aqueous battery electrolytesMeasured acceleration
- CheXzero tests label-free chest X-ray pathology detectionDirect discovery
- ProteinMPNN accelerates structure-conditioned protein sequence designMeasured acceleration
- NIH launches Bridge2AI for AI-ready biomedical research dataAdoption context
- AI infers candidate state variables from videos of physical systemsDirect discovery
- OC22 expands open catalyst data to oxide electrocatalystsInfrastructure
- National Academies frames automated workflows as science-speed infrastructureMeasured acceleration
- Deep reinforcement learning controls plasma shapes in a research tokamakMeasured acceleration
- FlyWire scales AI-assisted proofreading for whole-brain connectomicsDirect discovery
- DM21 uses deep learning to improve density-functional simulationsCapability benchmark
- Machine learning guides new conjectures in pure mathematicsDirect discovery
- EVE predicts disease variant effects from evolutionary sequence dataCapability benchmark
- Enformer improves DNA sequence-to-expression predictionCapability benchmark
- miniF2F standardizes formal Olympiad-level mathematics benchmarksCapability benchmark
- AlphaFold makes protein structure prediction a canonical AI-for-science milestoneLandmark discovery claim
- DINGO uses neural posterior estimation for real-time gravitational-wave analysisMeasured acceleration
- Physics-informed machine learning becomes a framework for scientific modellingLimitation or failure
- Broad launches a Schmidt-funded center for biology and machine learningAdoption context
- Therapeutics Data Commons standardizes AI tasks for drug developmentCapability benchmark
- cryoDRGN uses neural networks to model cryo-EM heterogeneityInfrastructure
- Bayesian optimization is benchmarked for synthetic reaction tuningCapability benchmark
- The Ramanujan Machine generates conjectures about mathematical constantsDirect discovery
- LHC Olympics benchmarks anomaly detection for collider physicsCapability benchmark
- nnU-Net automates configuration for biomedical image segmentationCapability benchmark
- CAMEO closes the loop for active-learning materials discoveryMeasured acceleration
- Fourier Neural Operator learns solution maps for families of PDEsInfrastructure
- MeshGraphNets extend learned simulation to adaptive mesh systemsMeasured acceleration
- CAMELS builds a simulation suite for machine-learning cosmologyDirect discovery
- A mobile robotic chemist autonomously searches photocatalyst mixturesMeasured acceleration
- Active machine learning helps identify CO2 electrocatalystsDirect discovery
- SciFact benchmarks evidence-based verification of scientific claimsCapability benchmark
- SPECTER uses citation-informed transformers for scientific paper embeddingsCapability benchmark
- Equivariant normalizing flows improve sampling for lattice gauge theoryDirect discovery
- Graph Network-based Simulators learn complex physical dynamicsInfrastructure
- Deep learning identifies halicin as an antibiotic candidateDirect discovery
- OmniFold applies machine learning to full-phase-space collider unfoldingCapability benchmark
- S2ORC provides machine-readable full text for AI literature toolsCapability benchmark
- FermiNet applies deep neural networks to many-electron wavefunctionsCapability benchmark
- AI Feynman uses physics-inspired symbolic regression to recover equationsCapability benchmark
- Crystal graph neural networks predict materials properties from structureInfrastructure
- Neural-symbolic AI plans organic synthesis routesCapability benchmark
- CaloGAN explores generative models for fast particle-detector simulationCapability benchmark
- Deep learning helps identify new exoplanets in Kepler dataDirect discovery
- Neural quantum states apply machine learning to many-body wavefunctionsInfrastructure