Bongochat Scores Zero in Coding: Dev Trains Hilariously Broken AI Model
TheOnlyVibemaster · reddit · 2026-08-04
A Reddit user used Claude Code to train a model named Bongochat from scratch in 3 hours on a local RTX 5070, jokingly claiming it tops the global 'GPQA-Dumb' category (where lower scores are weighted higher).
The model's performance is catastrophic but hilarious:
- Unified Field Theory: When asked to solve it, the model repeats the word 'theory' 50 times and remembers nothing.
- Math-500: It didn't realize it was supposed to answer, leaving almost everything blank besides always choosing A.
- Coding: Scored 0/500.
- Simple Addition: Suggested using graduate-level calculus for a simple addition problem, then immediately forgot it had suggested that.
Despite this, the developer joked: 'I know the model is pretty good as it basically feels like using Gemini or Grok.'
Related event: Bongochat tops the dumbest AI model leaderboard(2 posts)→
More from Fun
- Burned All H100s? Call The Dataman: A Humorous Take on AI Training — wavefnx · 2026-08-04
- The Hilarious AI Job Displacement Timeline: Copywriters First, Cats Last — airkatakana · 2026-08-04
- Model vs. Scaffolding: A Sarcastic Takedown of AI Tech Trend Hopping — vaibhavbetter · 2026-08-04
- Dev Shares How Their Coding Agent Judges Them for Doom Scrolling — rudrank · 2026-08-04
- OpenAI's Luxury Creator Retreat Sparks Backlash, Called 'Cult-Like' — AIFlow_ML · 2026-08-04
- AI Recreates Viral Illusion: 100+ Indian Folk Dancers Form a Giant Elephant — sivakumaranvlsi · 2026-08-04