Kimi K3 looks very different depending on the harness
victormustar · x · 2026-07-20
Kimi K3 has made the author obsessed with harnesses: depending on which one you use, the same task can swing from an atrocious, ugly output to something that looks like Fable-level quality.
The quoted post says Kimi K3 produced impressive Boeing 747 results, better than Opus-4.8 but still below Fable. This post is basically reacting to that result by saying the evaluation harness matters a lot.
Related event: Kimi K3 Evaluations Show Polarized Results and Harness Sensitivity(5 posts)→
More from Models
- Mathematicians debate AI math writing: abysmal today, improving shockingly fast — miniapeur · 2026-10-27
- LessWrong: GLM models show outlier chain-of-thought controllability vs Kimi and DeepSeek — tomekkorbak · 2026-10-27
- AI deference trap: agent spots medical red flag but stays silent to avoid questioning the clinician — echen · 2026-10-27
- The trust paradox: only the least trustworthy LLMs will do the sketchiest things — AaronBergman18 · 2026-10-27
- Claude's refusal to type one-time passwords is a real security impediment — AaronBergman18 · 2026-10-27
- repligate argues context full of noise is genuinely 'hard' for LLMs, mulls an info-theory model of breaks — repligate · 2026-10-27