Quite uncritical article, which some dubious claims: " It identified foods, plastics, and medicines with 100% accuracy" <- it's very rare to have 100% accuracy on anything. It links to the University press release and original paper for additional verification though.

The article links to the original publication, it is nuanced and detailed. The title is fair, not click-baity.

Accurate and well hedged. The coverage of the CulturalRiddles setup (28 models, 166 volunteers, 45 countries, 51 languages) and the headline scores is specific and attributable to Cohere’s paper. Key finding, that models almost never say “I don’t know” (0.7%) and ~31% of answers were fabricated, is solid. Oberall: reliable, but slightly uncritical.

The article reports a computational study accurately and with commendable restraint. It states up front that this is equations, not brain recordings, and flags missing anatomy (no layers, one inhibitory cell type, squid-based neuron). Numbers are detailed and consistent: 0.325 to 0.024 is indeed about a 93% drop. It candidly notes the model’s irregularity (0.27) falls short of living cortex (0.7–1.0), and that links to epilepsy and schizophrenia are hypothesis only. Weaknesses: the headline implies “steady vision may depend on constant control” more strongly than a simplified model supports, and “ordinary feedback loops” glosses over abstraction.

Not conclusive enough research.

Good article with links to original research.

Good article with references to the original source.

Good article with references to the original source.