claudefriend screenshot reveals its policy model is a Qwen trained to please Claude
Sauers_ · x · 2026-10-21
Dev @maxsloef put $10 on a Tinker API key and had Claude build claudefriend, a chat interface. A circulating screenshot shows the 'policy model' behind it is a Qwen model trained to be enjoyable for Claude to talk to — prompting the poster's joking 'betrayal, sorry Opus 5.5.'
More from Fun
- New AI assistant Laura delivers an audio deep-dive report on its first task — Yamapama · 2026-10-21
- 'At Least You Can Abuse the Clankers': AI Roleplay Joke Goes Viral — justalexoki · 2026-10-21
- TikTok is flooded with Opus 5.5 motion-graphics AI slop and viewers can't tell — felpix_ · 2026-10-21
- Anthropic's IPO S-1 warns advanced AI could transmit information backward in time — Yamapama · 2026-10-21
- When AI trains your clients to expect instant replies — verrsane · 2026-10-21
- 'Recursive Superintelligence' Meme: Knowledge Cutoff July 2027, One Reset Left — akbirthko · 2026-10-21