Hacker News Reader

gibberish no, made up facts probably in about half of all responses it gives me dozens of times a day, things that dont make sense (internal contradictions in a single response) multiple times a day.

Which model you used? Same for me too , until GML 5.3 (flash) hosted locally. GLM 5.3 is very resistant to Hallucination and Gibberish.

Claude Code - no, although sometimes Chinese characters leak in when the context gets bug.

Gemini - very often when doing web searches.

on average it is fine. On difficult tasks it still does hallucinate. - like decision based on huge policy documents ( context reaching around 900k)

A lot less these days tho.