Why ChatGPT Goes Insane (Anthropic research) скачать в хорошем качестве

Why ChatGPT Goes Insane (Anthropic research)

I am sure this has happened to you - you're mid conversation with ChatGPT, everything's fine, and then out of nowhere it starts acting completely different. Overly cautious, weird, toned down, and almost unrecognizable. Researchers at Anthropic also noticed the same issue, and they ran an experiment to figure out why this sudden drift occurs. They started by mapping 275 different AI personas inside 3 smaller models and discovered the assistant Axis. It's a measurable direction inside the model that tracks how "assistant-like" the model is behaving at any given moment. And certain types of conversations drag the model away from that axis mid conversation. This video breaks down the full research, what causes the drift, and how they're trying to fix it. Anthropic paper: https://arxiv.org/pdf/2601.10387 Thank you for watching, please leave a like if you learned something new. My website : https://www.parthknowsai.com/ Instagram:   / parthknowsai   tiktok:   / parthknowsai   linkedin:   / parth-patel-093652181