@thinkwithv: Ask Claude a yes/no question. Get a three-paragraph essay. Same?? Comment ADHD and I'll send you the fix 👇 You're not crazy — it's called length bias. These models learn from human ratings, and people rate longer answers as "better," so Claude pads to score higher. (Researchers call it reward hacking.) A 2023 model rambles less than today's. It's training, not intelligence. #claude #aislop

V | AI for Thinkers

Open In TikTok:

Region: US

Tuesday 04 August 2026 14:00:01 GMT