Does Claude perform better in 'claudish'? Researchers call for empirical measures
voooooogel · x · 2026-08-21
Discussion around Claude's increasingly distinctive "claudish" register: wassname asks whether Claude actually performs better when working in its own idiom — and would like to see empirical measures.
voooooogel doubts it, arguing any effect would be subtle and hard to measure, like asking whether a bilingual person works better in one language; RL pushes the model toward this register but doesn't destroy comprehension of regular English. He raises a converse concern: if Claude sometimes can't write plain English and needs an external humanizer model, off-policy text and lost introspection could degrade performance.
More from Models
- Zhipu's SAO: single-rollout async RL trains stably for 1,000 steps, beats GRPO — teortaxesTex · 2026-08-21
- Gemini's Safety Filters Too Strict? Rejects Kissing and Roadside Photos — Dry-Sympathy-3182 · 2026-08-21
- 0.63M Parameter Verifier Matches 7B Models in Specific Tasks — jm_alexia · 2026-08-21
- Pangram v4 Model Claims to Remove AI Watermarks and Mimic Human Writing — Scobleizer · 2026-08-21
- Brundage: something funky going on with ChatGPT inference, likely testing — Miles_Brundage · 2026-08-21
- What's left for hobbyists to post-train on in 2026? One bets on self-play poker — No-Compote-6794 · 2026-08-21