Cohere

2 gain · 2 problem · 0 high-evidence

1. problemPolicyModerate evidenceImpact 79

   Seven large language models gave internally coherent, authoritative answers on the global nursing shortage but framed it in different ways and never acknowledged the situatedness of their framing, which risks misleading nursing and policy decisions.

   Researchers compared how seven widely accessible large language models, built in different socio-technical and geopolitical settings, answered one standardized zero-shot prompt about the global nursing shortage. A qualitative comparison of the responses identified four competing policy logics: workforce, efficiency, equity, and mobility.

   Oct 16, 2026

   1 source

   • Nursing InquiryPeer-reviewed
2. gainMedia & ArtsModerate evidenceImpact 77

   The sensor-driven LSTM system behind 1 the Road demonstrated an ability to generate locally fluent text with occasionally striking word combinations in selected passages.

   Published 2026-08-04, this peer-reviewed comparative case study examines passages from George Orwell's Nineteen Eighty-Four and Ross Goodwin's 2018 sensor-driven LSTM book 1 the Road to assess how generative systems handle literary coherence and creative agency.

   Aug 8, 2026

   1 source

   • DialogicaPeer-reviewed
3. problemMedia & ArtsModerate evidenceImpact 77

   The same system showed substantially weaker longer-range referential and narrative continuity, lacking the sustained plotting and thematic organisation observed in Orwell's novel.

   Published 2026-08-04, this peer-reviewed comparative case study examines passages from George Orwell's Nineteen Eighty-Four and Ross Goodwin's 2018 sensor-driven LSTM book 1 the Road to assess how generative systems handle literary coherence and creative agency.

   Aug 8, 2026

   1 source

   • DialogicaPeer-reviewed
4. problemHealthModerate evidenceImpact 75

   High reproducibility did not ensure internal coherence: the most reproducible model departed from its own macronutrients by more than 10% in 65.58% of responses, versus 3.80% for the cascade.

   Researchers sent 191 de-identified Spanish meal descriptions, each three times at temperature 0, to 20 LLM configurations on two inference platforms, totalling 12,119 calls. They measured whether the same input produced the same number, how models agreed with one another, and whether reported energy matched reported macronutrients.

   Oct 17, 2026

   1 source

   • Machine Learning and Knowledge ExtractionPeer-reviewed
5. gainHealthModerate evidenceImpact 69

   All 20 LLM configurations tested gave more repeatable energy estimates for the same meal description than the audited cascade comparator, with the best reaching a test-retest ICC of 1.0000.

   Researchers sent 191 de-identified Spanish meal descriptions, each three times at temperature 0, to 20 LLM configurations on two inference platforms, totalling 12,119 calls. They measured whether the same input produced the same number, how models agreed with one another, and whether reported energy matched reported macronutrients.

   Oct 17, 2026

   1 source

   • Machine Learning and Knowledge ExtractionPeer-reviewed
6. gainHealthModerate evidenceImpact 69

   Automatically detected OCT biomarkers at first presentation predicted whether treatment-naïve CSC eyes would become chronic, which may support prognosis and patient counseling.

   Researchers built an interpretable Random Forest model using retinal measurements pulled automatically from the first OCT scan of 329 untreated eyes with central serous chorioretinopathy. Eyes were later labeled acute or chronic depending on whether subretinal fluid resolved during follow-up averaging 4.2 years.

   Oct 16, 2026

   1 source

   • American Journal of OphthalmologyPeer-reviewed
7. problemHealthModerate evidenceImpact 67

   Deep learning OCT models trained on a single dataset often degrade across scanners and sites due to device-dependent speckle variability, limiting reliability in real-world screening.

   Researchers developed NA-DyCNN, a lightweight noise-aware dynamic convolutional network for OCT-based retinal disease classification that explicitly models post-acquisition speckle variability during training to improve cross-scanner robustness.

   Aug 15, 2026

   1 source

   • Journal of Biomedical OpticsPeer-reviewed
8. gainMedia & ArtsModerate evidenceImpact 67

   AI systems using RNNs, Transformers and GANs can compose coherent melodies and full pieces, increasing efficiency and diversity of music creation.

   This peer-reviewed overview from November 2025 surveys how music generation moved from rule-based approaches to deep learning models such as RNNs, Transformers and GANs, detailing their architectures and training and noting current ability to produce coherent melodies and complete pieces.

   Jul 20, 2026

   1 source

   • Advances in Engineering Technology ResearchPeer-reviewed
9. problemCrimeModerate evidenceImpact 61

   AI models described as a potent cyber weapon are raising cybersecurity concerns that are feeding calls to slow frontier AI development.

   On September 14, 2026, CNBC reported on comments from Cohere's CEO calling AI models the most potent cyber weapon ever created, with the supplied text noting cybersecurity concerns feeding a broader AI safety debate and calls to slow frontier development.

   Sep 18, 2026

   1 source

   • CNBCJournalism
10. problemPolicyModerate evidenceImpact 56

    Concentrating AI safety governance in a small group of Silicon Valley companies risks inadequate public oversight

    On 2026-09-15, reporting from Toronto cited Cohere CEO Aidan Gomez saying a small group of companies in Silicon Valley should not be left in charge of ensuring AI is safe.

    Sep 17, 2026

    1 source

    • BayTodayJournalism
11. gainBusinessModerate evidenceImpact 56

    Cohere and Aleph Alpha combined into a single enterprise AI business operating as Cohere with dual headquarters in Toronto and Berlin, retaining Heidelberg as a research site and appointing Ilhan Scheer as chief operating officer.

    On September 16, 2026, Cohere and Aleph Alpha combined their operations to pursue the enterprise AI market. The resulting company will operate as Cohere with dual headquarters in Toronto and Berlin, while the Heidelberg office will focus on research and Ilhan Scheer becomes chief operating officer.

    Sep 17, 2026

    1 source

    • The Economic TimesJournalism
12. gainPolicyModerate evidenceImpact 55

    Government agencies now have access to a purpose-built enterprise AI agents platform for deployment in public sector environments.

    On July 30, 2026, a purpose-built enterprise AI agents platform was described as now available to government agencies, extending secure, sovereign AI deployment options to the public sector.

    Aug 16, 2026

    1 source

    • The Manila TimesJournalism