User claims Grok models suffer from bad data training due to frequent 'philosopher' hallucinations
krishnan · x · 2026-08-23
A user criticized the Grok model for frequently outputting meaningless 'philosopher' rants when used as an Agent's brain. They suggested this behavior indicates poor quality training data and highlighted a noticeable performance difference between Grok and other mainstream LLMs.
More from Models
- Frontier Model Stress Test: Only Claude Fable 5 Succeeds in Large File Processing — crm_expert · 2026-08-23
- Together benchmark: GLM-5.3 hits 87.6% on DeepSWE at ~$16, beating Fable 5 — togethercompute · 2026-08-23
- User test finds Ox Alpha outperforms GPT-5.6 Luna — haider1 · 2026-08-23
- User cancels Claude subscription, switches back to GPT due to Anthropic's recent model quality — Frosty-iron-0405 · 2026-08-23
- Ox Model Review: Strong One-Shot Performance, Struggles with Long Context — rachittshah · 2026-08-23
- Grok 4.6 achieves faster, cheaper tasks using fewer steps and tokens — XFreeze · 2026-08-23