A dataset from Meta FAIR, the University of Toronto, and VSParticle covers over 600 mixed-metal catalysts made from 13 elements and tested under the same conditions, alongside analysis of over 19,000 materials. Hydrogen-evolution predictions flagged platinum-containing alloys as promising, although platinum was absent from training data. Fewer than 25% of produced catalysts matched the target; this was a pilot.
Technology: Catalyst prediction models (Open Catalyst)
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: Computational predictions were compared with lab results: 441 gas diffusion electrodes tested in zero-gap electrolysis and 572 characterized samples, alongside DFT-verified adsorption energies on about 20,000 materials.
- Still unknown: Platinum was ranked as a top hydrogen-evolution candidate but was not experimentally tested, and the abstract lists no explicit limitations.
- AI role: AI-accelerated relaxations (685 million) fed a data-driven Sabatier volcano that flagged platinum-containing alloys as promising.
Evidence sources: https://arxiv.org/abs/2411.11783
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
PatternBoost alternates a local search that generates candidate constructions with a transformer trained on the best ones. The authors report the best known solutions to several extremal combinatorics problems, including a counterexample to a conjecture that had been open for 30 years. Candidates are scored mechanically, so each construction is checked by computer; the abstract does not name the conjecture and the paper was not peer-reviewed at the time of posting.
Technology: PatternBoost (transformer trained on search outputs, combined with local search)
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- Tested: The abstract reports best known solutions to several extremal combinatorics problems, including a counterexample to a 30-year-old conjecture, but describes no verification procedure.
- Still unknown: The abstract states that PatternBoost's performance varies across problems.
- AI role: A local search generated candidate constructions, and a transformer trained on the best candidates seeded the next search round.
Evidence sources: https://arxiv.org/abs/2411.00566
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
A three-part system (image encoder, brain encoder, decoder) reconstructs viewed images from magnetoencephalography signals at each time point, with signals better aligned to DINOv2 than the other models compared. Images remain imperfect; fMRI decodes better; the page gives no numerical metrics; preprint.
Technology: Magnetoencephalography decoder
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- Tested: MEG recordings from healthy volunteers viewing images were decoded at each time point; several pretrained image models were compared (signals aligned best with DINOv2), and MEG decoding was compared with fMRI. Participant and image counts not stated on the blog.
- Still unknown: Generated images are imperfect: high-level features are often kept but low-level features are often wrong. MEG decoding is less precise than fMRI decoding.
- AI role: AI models built the image representations and generated the output images; volunteers supplied the brain recordings.
Evidence sources: https://ai.meta.com/blog/brain-ai-image-decoding-meg-magnetoencephalography/ · https://arxiv.org/abs/2310.19812
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Meta AI published predicted structures for almost the entire MGnify90 database (over 600 million), generated in about two weeks on roughly 2,000 GPUs with a 15-billion-parameter language model, up to 60 times faster than previous methods. These are predictions, and high-confidence structures are a much smaller subset.
Technology: ESMFold (15-billion-parameter language model)
Family ESM / ESMFold: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- Tested: The preprint reports structure prediction up to 60 times faster while maintaining accuracy versus state-of-the-art methods and an atlas of 617 million structures; the blog describes no experimental checks.
- Still unknown: The blog lists no limitations and notes some sequences fold into disordered forms; high-confidence predictions are a subset.
- AI role: Language model (15B parameters) predicted protein structures from sequence.
Evidence sources: https://ai.meta.com/blog/protein-folding-esmfold-metagenomics/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1101/2022.07.20.500902&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.