"Someone should train LLMs to approximate human writing distribution" backfires
AaronBergman18 · x · 2026-10-15
A post complaining that "LLMs are so bad at writing" and calling for "a training objective that makes them approximate the distribution of human writing, even before post-training" went viral as an AI-community joke.
The punchline: that objective already exists — it's essentially next-token pretraining itself. The critic unknowingly proposed the very paradigm behind LLMs, and the irony is what made it shareable.
More from Fun
- Hackathon team claims it found a large fleet of agents from a non-US frontier lab — NathanpmYoung · 2026-10-15
- Robots still need a bigger poop bucket, per robotics insider — justalexoki · 2026-10-15
- Internal OpenAI meme: revenues in the mirror are closer than they appear — charles_irl · 2026-10-15
- A Bradbury short story about creating one real thing, retold — moultano · 2026-10-15
- A Banana Lawyer Demands Justice: Nano Banana Courtroom Video Goes Viral — NegotiationSmooth467 · 2026-10-15
- 'The Telescope Made Astronomers Obsolete': A Sarcastic Jab at AI Job Panic — littmath · 2026-10-15