Really wish there was a way to study the training data and processes of chatgpt other than by inference. Which is to say, I wish these people would stop intentionally blackboxing their damn systems so they could be investigated by sincere researchers.
Like… if these are real, then this OpenAI chatGPT/Microsoft Bing integration is… Really not ready for public use. I mean, these examples legit read like you gave a recalcitrant 4 y.o. epistemic and motivational determinative authority.
Which, depending on the nature and persistence of these errors, could turn genuinely deadly.
https://www.theverge.com/2023/2/15/23599072/microsoft-ai-bing-personality-conversations-spy-employees-webcams
Agreed, if scientists could study what it was doing then they could come up with ways to try and improve what it is doing.
