Hypothesis on Opus 5 leakiness: Mixed old and new training formats
Ratter · x · 2026-09-01
A user hypothesizes that Opus 5's tendency to 'leak' (in fakebase mode) is due to training on chatlogs in both old and new formats.
It is observed that when continuing in fakebase mode, the model sometimes uses old \n\nHuman: / \n\nAssistant: markers, or shorthand like 'H'/'A', and sometimes even 'user' without a space.
Related event: Users Say Opus Training Shift to XML Tags Explains Leak Behavior(3 posts)→
More from Models
- OpenAI Codex sees exponential growth amid Anthropic user decline — FinanceYF5 · 2026-09-01
- Meme Suggests AI Models Are Distilling Each Other — toomanynamesaretook · 2026-09-01
- Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored Trends on HF — DavidAU · 2026-09-01
- Matrix-Game 3.5: Enhancing Real-Time Streaming Interactive World Models with Patch Memory — Runjia Qian · 2026-09-01
- Qwen Team Analyzes Qwen3.8-Next Architecture Design — Qwen · 2026-09-01
- Zhipu Releases INT4 & MXFP4 Versions of GLM-5.3 Flash — HaihaoShen · 2026-09-01