From the archive
Wednesday, 28 October
Folded from 11 sources that day.
Sorted by quality and by how many sources covered it, the same for everyone.
-
Gemini 4 Carbon: Our next frontier model
Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index at 59 and leads several cybersecurity benchmarks. Access starts with trusted defenders through the Fairwind Program before broader release.
-
The Benchmarking Crisis
Argues that AI-driven optimisation is overfitting database benchmarks like ClickBench, with examples of hash-only grouping and cached predicates that are fast but not generally valid.
-
OpenAI and ServiceNow expand partnership across enterprise workflows
ServiceNow will bring OpenAI models and agents into its products and connect its workflows to ChatGPT and Codex. Planned features include a Codex integration for app development and a joint cybersecurity solution.
-
Artificial Analysis v5 and the benchmarker's dilemma
Artificial Analysis's Intelligence Index v5 weights knowledge work, coding, science and general ability at 25% each. The author argues that single-number AI benchmarks are useful but deeply flawed.
-
Do We Still Need Data Models?
Argues that agents may take over much of data system design, but models remain necessary to convey what the system means and which interpretations and decisions were made.