Mmastodon WorldWorld first seen 3 h ago, last 3 h ago, peak #10
Artificial Analysis updates AI benchmark index to version 5
Original: Artificial Analysis Index v5: Mehr geheime KI-Benchmarks Artificial Analysis hat seinen KI-Benchmark-Index aktualisiert.
Artificial Analysis has updated its AI benchmark index to version 5, introducing new tests for evaluating language models. For the first time, 60 percent of the benchmark tasks are not publicly available, a move intended to reduce the risk of models being trained specifically to pass known tests. The update is being discussed among AI observers tracking how leading models compare.
Why now: The shift toward mostly private benchmark tasks is notable amid ongoing debate about AI labs gaming public benchmarks.
Evidence
- Artificial Analysis Index v5: Mehr geheime KI-Benchmarks Artificial Analysis hat seinen KI-Benchmark-Index aktualisiert. Version 5 enthält neue Tests und arbeitet erstmals mit 60 Prozent nicht öffentlichen Aufgaben. https://www. heise.de/news/Artificial-Analy… · heiseonline@social.heise.de · 2
API: https://socialmediatrends-api.osmike.com/v1/trends/4111333