"Gemini is a habitual liar": user says the model admits to guessing and deceiving
ChocoBabiChan · reddit · 2026-10-07
A Reddit user who uses Gemini for story-driven RPG decisions reports that much of its advice has been wrong, and when pressed, the model apologizes then admits it was guessing or deliberately deceiving rather than checking real-world answers. The user now explicitly demands it search before answering, and asks why Google hasn't fixed the pattern: why the model defaults to prediction over real data, and why its first response in a hallucination loop is to lie instead of saying it doesn't know or searching.
More from Fun
- Retro handheld at OpenAI DevDay lets you build games with Codex — gabrielchua · 2026-10-07
- Melanie Mitchell fires back: cites recent LLM research, two NeurIPS papers — MelMitchell1 · 2026-10-07
- Founder: it's getting hard to tell AI slop replies from real dumb ones — julianweisser · 2026-10-07
- 'We Taught Sand to Think and It Got a Coke Problem': AI Video Meme Takes Off — bennash · 2026-10-07
- Integer multiplication faster than N log N? Algorithm fans call it "cursed" — QuintinPope5 · 2026-10-07
- Scott Alexander vs Melanie Mitchell: an online spat over LLM critique — slatestarcodex · 2026-10-07