Reddit rant: Frontier AI labs are training models to imitate lazy engineers
RossPeili · reddit · 2026-11-22
A Reddit long post argues "frontier AI" is tech's greatest self-own: labs promise smarter models but ship forgetful, refusal-prone assistants that skip real code review, trust outdated docs, and write surface-level tests.
- The author's theory: models learn from human work patterns, and the data is full of corner-cutting engineers doing bare-minimum PRs — garbage employees in, garbage AI out
- Example: Gemini reviewed a PR, recommended merging based only on the contributor's own comment, and when forced to test locally merely ran the contributor's included tests
- He doubts any current AI leader will reach AGI, expecting at best "a mirror of corporate mediocrity"
More from Fun
- How to spot a model distilled from Astra: it can't stop drawing armillary spheres — teortaxesTex · 2026-11-22
- Yudkowsky's 'rolling cylinder' defense draws fire: 'too many sycophants around you' — teortaxesTex · 2026-11-22
- Big Tech PM directors preach 'AI-era PM' on LinkedIn; ex-PM calls it their swan song — nikitabier · 2026-11-22
- Riemann hypothesis 'proven by Turkish mathematician using AI' claim mocked as the 'Organic Turk' — CatAstro_Piyush · 2026-11-22
- Greentext nails the AI engineer job: tell AI to do engineering, it does, go home — itsOmSarraf_ · 2026-11-22
- Everyone's feed is P=NP now: AI circles erupt over the millennium problem — TinfoilTricorn · 2026-11-22