Physicist Sabine Hossenfelder: LLMs Still Fail at Writing Video Scripts
skdh · x · 2026-08-01
Physicist and YouTuber Sabine Hossenfelder shared her long-term experience using ChatGPT, Claude, Grok, and Gemini for YouTube video scripts, concluding they are a "complete failure."
She highlighted several core issues:
- Stale Topics: Models inevitably suggest widely discussed and unoriginal ideas.
- Incoherent Logic: Even with a specific topic, the generated scripts are incomprehensible, repetitive, and structurally poor, despite having coherent individual sentences.
- Poor Fact-Checking: Models fail to verify statements accurately and often flag correct information as wrong.
She expressed frustration that their ability in long-form logical writing has worsened over the past two years. Currently, she finds LLMs useful only for finding references and fixing English grammar.
Related event: Tech Blogger Tests Top LLMs for Video Scripts, Finds Them Failing(4 posts)→
More from Models
- DeepSeek V4 Impresses, Powering Bizarre Full-Moon Code Reset Study — Teknium · 2026-08-01
- Microsoft Teases 'Astra' Model: Solves 10 Complex Math Conjectures with Lean Proofs — ctjlewis · 2026-08-01
- Poolside Laguna S 2.1 Updates FP8 Weights with Native 1M Context — rmhubbert · 2026-08-01
- Yale and UChicago Study: LLMs Generate Narrower Research Ideas Than Humans — rohanpaul_ai · 2026-08-01
- Gemma 3 27B Fails at Local Coding: File Edits Break Due to Indentation Mismatches — DanTup · 2026-08-01
- DeepSeek Harness Enters Closed Beta, Recruiting Agent Open-Source Developers — jiayuan_jy · 2026-08-01