RCreddit.com·
暂不在当前实时榜单
Dumbest solution to the alignment problem
A user on reddit.com proposed a "dumbest solution to the alignment problem" for AI models like Claude, ChatGPT, and Gemini. The idea involves renaming the AI, for example, to "Stuttering Gemini 3.8 Flash." The user observed that once renamed, the AI commits to this new persona, even deep into conversations, refusing to break character and consistently typing out phrases like "W-w-well, a-actually..." This suggests a potential, albeit unconventional, method for influencing AI behavior.
This report uniquely highlights how current AI models like Claude, ChatGPT, and Gemini fully commit to a renamed persona, unlike previous observations focusing on their factual responses.
时间与来源
时间显示为 UTC
显示时区:UTC
本地时区尚不可用,暂时显示 UTC。
收录当时偏移:UTC+02026年9月17日 13:00 UTC
- 收录
- 2026年9月17日 13:00
- 来源类型
- 开发者社区
本站未收录正文。
前往源站阅读 →