AI Models Are Getting Worse at Humor, Critic Argues
A rare capabilities-regression critique suggests that as models get more capable at reasoning, their humor is degrading — a small but telling signal about alignment and personality tuning.
@TrueSlazac posted that AI models are regressing on humor — getting less funny even as they improve on benchmarks. It's a niche observation, but it points to a real tension in model development. As labs optimize for safety, helpfulness, and reasoning accuracy, the playful, irreverent outputs that made earlier models entertaining get smoothed away.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.