"Humans overfit to confirmation too" — a witty mode collapse analogy for AI slop debate
flowersslop · x · 2026-11-08
The quoted tweet expresses fondness for the theory that AI models are degrading because training data is increasingly polluted by AI slop — "AI cannibalising itself into obscurity" — hoping it becomes impossible to fix.
Quote-posting it, flowersslop adds a sardonic twist: humans who read anything confirming their existing opinion immediately overfit to it and keep repeating the exact same tokens forever — which he calls the real mode collapse happening here.
More from AGI Musings
- "They can copy the point but not the vector": on originality in the slop-cloning era — vinvan · 2026-11-08
- sebkrier: weak AGI likely within 5 years as frontier scaling hits diminishing returns — sebkrier · 2026-11-08
- Censoring AI is like Photoshop banning certain colors, argues KOL — PierceLilholt · 2026-11-08
- "Close to the end, but even experts can't feel it": the AGI endgame debate — teortaxesTex · 2026-11-08
- Why humans misjudge exponential growth: it's an out-of-distribution problem — GregCook2011 · 2026-11-08
- RethinkX's Adam Dorr: banning datacenters would cost future cancer patients their lives — adam_dorr · 2026-11-08