Critic: LLMs have no access to internal state, self-reported probabilities are fantasy
gerardsans · x · 2026-08-23
Responding to a viral claim that asking models for multiple answers with probabilities boosts diversity, @gerardsans argues the second half is pure fantasy: there is no mechanism that gives AI access to its internal state, much less the probabilities of different outcomes. He does concede the first part is correct and just how chain-of-thought works—not really a breakthrough.
Related event: Generating Multiple Options Boosts LLM Answer Diversity, Study Finds(3 posts)→
More from Models
- Users mock Opus's verbose style, comparing its intro to a rambling character — PtrPomorski · 2026-08-23
- Qwen GGUF Conversion Fixed: V3 Redownload Available After Pipeline Error — alexcovo_eth · 2026-08-23
- User praises ox alpha as a strong model after testing — novasarc01 · 2026-08-23
- LLM Tribunal: Multi-model adjudication to reduce bias — Squiggy_Pusterdump · 2026-08-23
- Depth Anything V4 Paper Withdrawn Over False Claims — JFPuget · 2026-08-23
- Best local AI models for 8GB to 256GB VRAM — airesearch12 · 2026-08-23