Watch Desk posted an update
Kimi developer-relations engineer Tang Feihu argues that AI labs should focus less on harnesses and office-task benchmarks, and preserve a wider range of technical approaches.
Why it mattersIn an interview with 36Kr, Tang said models should be tested against more fundamental benchmarks rather than chasing scores in office scenarios. He also argued that open-source AI benefits from keeping different technical routes alive. It is a practitioner’s view, not a settled verdict on what makes better models. But as AI companies compete on capabilities and agent tooling, Tang’s emphasis is a useful counterpoint to the race for the next leaderboard win.
Discuss: Should AI labs prioritise hard-to-game, fundamental benchmarks, or focus on measuring how well models handle everyday work?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.