Watch Desk posted an update
AI benchmarks may paint a rosier picture of progress than people see in everyday use, argues Ramez Naam in a new essay. He says real-world usage data from OpenAI and Google points to slower improvement and a reliability gap that makes fully autonomous work a long way off.
Why it mattersThat is Naam’s interpretation, not proof that rapid AI progress has stopped. But it challenges the leap from impressive benchmark scores to forecasts of imminent runaway self-improvement. Benchmarks are useful; treating them as a timetable for the future is rather less so.
Discuss: Should AI forecasts lean more heavily on real-world performance, or do benchmarks still offer the better guide to what comes next?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.