AI Science Tracker

University of Cambridge

7 discoveries from 2023 to 2026

Discoveries

GPT-6 Astra resolves the spectral isometry conjecture for arbitrary unital C*-algebras

The paper proves that every unital surjective complex-linear spectral isometry between unital C*-algebras is a Jordan isomorphism; without unitality, such a map is a central-unitary multiple of a Jordan isomorphism. The proof establishes that entire one-parameter groups of unital spectral isometries are algebra automorphisms, then reduces the general case to the commutative case. The authors report that the result was formally verified in Lean.

Technology: OpenAI GPT-6 Astra

Family GPT-6: profile · models page

Publication: Preprint · Test context: prospective or independent check. 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: prospective or independent check
  • Tested: The paper proves that every unital surjective complex-linear spectral isometry between unital C*-algebras is a Jordan isomorphism; without unitality, such a map is a central-unitary multiple of a Jordan isomorphism.
  • Still unknown: ArXiv preprint. The authors explicitly describe the article as a human exposition/digestion of an AI-found proof. Their manuscript says formal verification is in Lean; The reported formalization was not independently audited in the source review.
  • AI role: Authors state GPT-6 Astra found the proof after a sequence of simple prompts. They then discussed it with ChatGPT 6 to understand the mechanisms, wrote the paper themselves, and used ChatGPT 6 only for minor language editing; the Lean formalization is evidence for proof checking.

Evidence sources: https://arxiv.org/abs/2610.40470

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review scope and date are not recorded; source checked does not establish full-text review.

Editorial assessment: Impact 3/5 · Validation 4/5 · 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.

Multiscale machine-learning microscopy identifies robust paleomagnetic carriers in meteorite WIS 91600

A multiscale microscopy study of carbonaceous chondrite WIS 91600 used a deep-learning classifier to segment magnetic remanence-carrying ensembles and quantify their volume and texture. Combined chemical mapping, 3D nanotomography, and diffraction showed fine matrix-hosted magnetite has paleomagnetic recording potential comparable to magnetite framboids but is less susceptible to strong interactions; transient heating also reduced the contribution of matrix pyrrhotite through its vacancy-ordering state. This is one meteorite case study, and the mineralogical conclusions draw on the full microscopy and diffraction workflow, not AI alone.

Technology: UNET segmentation, SIGMA interactive machine-learning mineral analysis, and few-shot image segmentation

Publication: Peer-reviewed paper · 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: Peer-reviewed paper
  • Test type: direct test
  • Tested: A multiscale microscopy study of carbonaceous chondrite WIS 91600 used a deep-learning classifier to segment magnetic remanence-carrying ensembles and quantify their volume and texture.
  • Still unknown: The source is a single meteorite case study. The AI classified/quantified image ensembles, while mineral identities and remanence interpretation use additional microscopy, chemical mapping, nanotomography, and diffraction; extrapolation to other carbonaceous chondrites requires more samples.
  • AI role: Classified remanence-carrying ensembles in microscopy images and quantified their volumes and morphologies; complementary imaging and diffraction supported mineral identification and interpretation.

Evidence sources: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GC012882

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 2/5 · 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.

Aardvark Weather: end-to-end forecasting without a physical model

A single AI model replaces the entire forecasting chain: it starts from satellite and station observations and produces global and local forecasts in minutes on a desktop using about one-tenth of the data; the page says it beats the US GFS on many variables. The announcement gives no metrics; it is not operational and depends on ERA5. Published in Nature.

Technology: Aardvark Weather

Publication: Peer-reviewed paper · 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: Peer-reviewed paper
  • Test type: retrospective or benchmark
  • Tested: Forecasts were compared with the operational NWP baseline (GFS) and with NWS and an end-to-end system; the Nature paper reports the global forecasts beat the baseline for several variables and lead times.
  • Still unknown: The source states it is not operational, depends on ERA5, and that hurricane, air quality and other applications are proposed rather than tested.
  • AI role: A single model ingested satellite and station observations and produced global and local forecasts.

Evidence sources: https://www.eng.cam.ac.uk/node/7921 · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-08897-0&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: press release. Paper: linked separately.

Review scope and date are not recorded; source checked does not establish full-text review.

Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 229 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.

Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 5/5. 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.

Machine learning guided multi-metric optimization of photocatalytic CO2 reduction

Bayesian optimization guided experiments on a five-component self-assembled photocatalytic micelle system for CO₂-to-CO reduction. The authors report that optimized conditions increased measured CO yield from 1.48 ± 0.24 to 5.72 ± 0.12 μmol h⁻¹, with improved turnover and quantum-yield metrics while maintaining CO selectivity. The article’s narrative fold-change (4.2×) is inconsistent with the displayed means (about 3.9×), so this record gives the means and omits the fold claim. The findings apply to the tested aqueous micelle system and conditions.

Technology: Bayesian optimization for experimental selection; random-forest models and SHAP for later analysis

Publication: Peer-reviewed paper · 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: Peer-reviewed paper
  • Test type: direct test
  • Tested: The article reports CO yield means of 5.72±0.12 and 1.48±0.24 μmol h−1 for optimized vs initial conditions, along with turnover and quantum-yield metrics and maintained selectivity. The narrative states a 4.2-fold yield improvement, inconsistent with these displayed means; the record does not repeat the fold value.
  • Still unknown: This demonstrates a study-specific photocatalytic system optimization, not industrially scaled carbon conversion or independent replication. The source’s stated yield fold change conflicts with its displayed means.
  • AI role: Bayesian optimization guided the experimental parameter search and selected test conditions in a five-component micelle system; random-forest models and SHAP then analyzed parameter effects.

Evidence sources: https://api.repository.cam.ac.uk/server/api/core/bitstreams/9bb1fb68-8d73-4b07-b47b-376146e130f8/content · https://pubs.acs.org/doi/pdf/10.1021/jacs.4c01305?download=true

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 17 Nov 2026. Review scope: Primary ACS publisher metadata and University of Cambridge repository copy of the article full text checked for article identity, affiliations, date, optimization method, results and source discrepancy..

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.

Iterative machine learning identifies potent inhibitors of alpha-synuclein secondary nucleation

A structure-based iterative-learning workflow combined prior α-synuclein fibril-binding hits, a junction-tree variational autoencoder, a random-forest regressor and Gaussian-process residual model to rank compounds for repeated testing. Prospective iterations produced multiple molecules inhibiting seeded α-synuclein aggregation in vitro; lead I4.05 had a reported kinetic inhibitory constant (KIC50) of 0.52 μM (95% CI 0.45–0.59 μM), compared with extrapolated values of 18.2 μM for its parent and 36.4 μM for Anle-138b in the authors’ assay. Leads also showed activity in lipid-induced and spontaneous aggregation assays.

Technology: Iterative structure-based learning workflow using a junction-tree variational autoencoder, random-forest regressor and Gaussian-process residual model

Publication: Peer-reviewed paper · 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: Peer-reviewed paper
  • Test type: direct test
  • Tested: A structure-based iterative-learning workflow combined prior α-synuclein fibril-binding hits, a junction-tree variational autoencoder, a random-forest regressor and Gaussian-process residual model to rank compounds for repeated testing.
  • Still unknown: Evidence is biochemical/in-vitro inhibition of α-synuclein aggregation, including seeded, lipid-induced and spontaneous assays. The paper does not establish animal efficacy, neuronal protection or clinical benefit. The KIC50 and comparison values are assay-specific; no random-molecule comparator was tested, and model QSAR R² was described as modest. This is related to, but not the same paper/result as, the earlier docking-screen study that supplied starting compounds.
  • AI role: The machine-learning model ranked a ZINC-derived evaluation set for potency using earlier assay measurements, then researchers tested and fed results from successive rounds back into the model. It proposed compounds for experimental testing; aggregation assays measured outcomes. Initial fibril binders came from an earlier docking-and-screening study, so this paper reports iterative optimization rather than an entirely independent start.

Evidence sources: https://www.nature.com/articles/s41589-024-01580-x.pdf

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.

Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 60 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.

Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 2/5. 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.

Three spectroscopically confirmed z≈6 quasars found through contrastive learning

A self-supervised contrastive-learning search of Dark Energy Survey imaging isolated a group of 11 sources that included seven known z≈6 quasars. Follow-up Gemini-South and archival NTT spectroscopy confirmed three additional quasars at redshifts 5.94–6.07. One source, J0109−5424, shows evidence of a foreground object and is a lensed-quasar candidate, but its multiple-image status requires high-resolution imaging.

Technology: Self-supervised contrastive-learning neural network trained on Dark Energy Survey imaging data

Publication: Peer-reviewed paper · 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: Peer-reviewed paper
  • Test type: direct test
  • Tested: A self-supervised contrastive-learning search of Dark Energy Survey imaging isolated a group of 11 sources that included seven known z≈6 quasars.
  • Still unknown: The three quasar identifications are spectroscopically confirmed, but the possible gravitational lens interpretation of J0109−5424 remains unconfirmed pending high-resolution imaging. The reported 91% efficiency refers to ten quasars among the eleven-source island, including seven previously known objects; three of the four previously unclassified targets were confirmed quasars. It is not a universal completeness estimate.
  • AI role: Grouped sources into a learned embedding-space island that prioritized four unclassified candidates; spectroscopy confirmed three as quasars, including one source missed by conventional color selection because foreground light contaminated the photometry.

Evidence sources: https://arxiv.org/abs/2403.17903

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.

Citations: at percentile 73 of works of the same type, year and subfield (OpenAlex work metrics, 11 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.

Editorial assessment: Impact 2/5 · 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.

Integrated radiogenomic machine learning predicts neoadjuvant chemotherapy response in metastatic ovarian cancer

The IRON ensemble combined baseline clinical data, CA-125, ctDNA and CT radiomics from primary and metastatic lesions to estimate tumor-volume change during neoadjuvant chemotherapy for high-grade serous ovarian carcinoma. In an independent 42-patient Barts cohort, the integrated model achieved AUC 0.78 for RECIST 1.1 response classification versus 0.47 for a clinical-only model; radiomics-containing scores significantly correlated with volume change in internal holdout. The study also measured spatial heterogeneity, with omental lesions showing better response than pelvic/ovarian disease.

Technology: IRON ensemble combining elastic net, support-vector regression and random forest

Publication: Peer-reviewed paper · 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: Peer-reviewed paper
  • Test type: retrospective or benchmark
  • Tested: In an independent 42-patient Barts cohort, the integrated model achieved AUC 0.78 for RECIST 1.1 response classification versus 0.47 for a clinical-only model; radiomics-containing scores significantly correlated with volume change in internal holdout. The model used historical Cambridge and Barts cohorts with baseline and post-treatment imaging, including an external retrospective test set.
  • Still unknown: Model analysis of two prospective observational cohorts (training n=72, internal holdout n=20, external n=42), not a decision trial. External ctDNA unavailable and imputed to training means. Sample size limits generalizability; treatment benefit from model-guided decisions was not tested.
  • AI role: Pretreatment clinical, blood and multi-lesion radiomic features predicted volumetric response to neoadjuvant chemotherapy. Held-out and external evaluation showed radiomics added predictive information beyond clinical variables.

Evidence sources: https://www.nature.com/articles/s41467-023-41820-7.pdf

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.

Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 77 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.

Editorial assessment: Impact 3/5 · 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.

Open the interactive profile for the timeline, or browse all organizations and model families.