Theo self-runs Terminal Bench 4, touts Opus-beating value
Theo said he ran Terminal Bench 4 himself because benchmarks were missing, claiming performance above Opus 5.5 at roughly one-thirtieth the price.
benchmark
All AI Pulse updates tagged "benchmark".
8 signalstimm adds multi-label classification support
Ross Wightman says timm now supports multi-label classification, polishing up an implementation from last year's Task refactor that remains useful as a benchmark.
fchollet cites ARC-AGI-(n+1) meta-benchmark
François Chollet suggests a test for human-level generality: passing ARC-AGI-(n+1) immediately upon release as a meta-benchmark.
Stanford debuts PhilosophyBench for AI philosophy
Sebastian Thrun introduces PhilosophyBench from Stanford AI Lab and Stanford HCI, billed as the first independent large-scale benchmark for AI philosophical capabilities.
SWE-Bench ProMax Benchmarking Agents on Large-Scale Multilingual Code Refactoring paper: http...
SWE-Bench ProMax Benchmarking Agents on Large-Scale Multilingual Code Refactoring paper: https:// huggingface.co/papers/2608.09 802 …
There is only one math benchmark left
There is only one math benchmark left.
Why does the bicycle scene have bad PSNR?
There is a dog in the dataset haunting your benchmark.
How the brain tracks internal state
Exploring how the brain builds and tracks an internal state.