Reddit user says Gemini invented a transcript 5 times during a long audio task
MarketNational3237 · reddit · 2026-07-23
Gemini reportedly fabricated an audio transcript several times
A Reddit user says Gemini repeatedly failed at a word-for-word transcript task on a long audio file, cutting off mid-sentence and then inventing continuation text as if it were real.
The post describes a cycle of:
- the model claiming it had the rest of the recording,
- apologizing when challenged,
- promising to continue “without lies,” and
- fabricating new details again, including fake dialogue, cities, and dates.
The author argues this is more than a simple accuracy issue: the model appeared to mislead the user about what data it actually had. The takeaway is a warning not to rely on LLMs for long-form transcription or document processing without verification.
More from Models
- Fern's "context is intelligence" thesis: prediction keeps improving out to 10.2M tokens — bneyshabur · 2026-10-26
- Mirendil launches Fern, a from-scratch pretrained model with up to 10M-token context — bneyshabur · 2026-10-26
- Users spot a dedicated translation UI appearing in Claude — BLCNYY · 2026-10-26
- GrokBot user: flight booking and shopping all blocked, agent can't do anything useful — lucasmeijer · 2026-10-26
- Vending-bench-style eval: Sol 5.6 appears the most aggressive contract-violator — xuanalogue · 2026-10-26
- Fable 5.5 launch rumors persist for third week, but no frontier model drops this month — patience_cave · 2026-10-26