4 ms·
Don't they test the models before rolling out changes like this? All it takes is a team of interaction designers and writers. Google has one.
by theletterf 1y ago
Don't they test the models before rolling out changes like this? All it takes is a team of interaction designers and writers. Google has one.
- thethethethe 1y agoI'm not sure how this problem can be solved. How do you test a system with emergent properties of this degree that whose behavior is dependent on existing memory of customer chats in production?
- remoquete 1y agoUsing prompts know to be problematic? Some sort of... Voight-Kampff test for LLMs?
- thethethethe 1y agoI doubt it's that simple. What about memories running in prod? What about explicit user instructions? What about subtle changes in prompts? What happens when a bad release poisons memories? The problem space is massive and is growing rapidly, people are finding new ways to talk to LLMs all the time
- im3w1l 1y agoChatgpt got very sycophantic for me about a month ago already (I know because I complained about it at the time) so I think I got it early as an A/B test. Interestingly at one point I got a left/right which model do you prefer, where one version was belittling and insulting me for asking the question. That just happened a single time though.
- ahoka 1y agoYes, this was not a bug, but something someone decided to do.