# Theoretical Limits on Hallucination

Formal learning theory increasingly frames hallucination as a bounded property rather than an eliminable defect, shifting reliability work toward verification layers around models.

- Conviction: 26 / 100 (weakening)
- 7-day move: -14
- Horizon: Emerging (watchlist)
- Tracking since: 2026-06-30T00:00:00.000Z
- Last updated: 2026-07-07T14:00:02.329Z
- Canonical: https://polylog.news/ai/trends/hallucination-theoretical-limits
- Publisher: Polylog
- Affected regions: Global

## Recent score history

- 2026-07-06: 28
- 2026-07-07: 26

## Recent evidence

- [confirms] A Theory Paper Asks Whether Language Models Can Learn While Hallucinating Forever (2026-06-30): A new theory paper extends the 'generation in the limit' framework to ask whether language models can keep learning while producing invalid strings forever, framing hallucination as a bounded formal property.