“…over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts

“…over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts as sensitive.” As predicted. Nobody did anything wrong. Google did great. Machines make mistakes. Trust the code. https://news.ycombinator.com/item?id=39466920 https://news.ycombinator.com/item?id=39466920