4 ms·
> What is the point of dosubot? It agrees, and apologizes for not having provided a supposedly adequate response. This is my problem with chat gpt - I can’t ha
by mock-possum 3y ago
> What is the point of dosubot? It agrees, and apologizes for not having provided a supposedly adequate response.
This is my problem with chat gpt - I can’t have anything resembling a ‘real’ conversation with the bot, because it refuses to engage with me in any way other than:
1) restate my last message
2) agree with what I’ve said to an excessive degree, possibly even complimenting me on having said it
3) ask me a leading question that encourages me to send another message, and now goto 1 and loop forever.
It ends up just feeling like I’m the target of some kind of time wasting prank.
- fhars 3y agoWell tested communication strategy, that is how Eliza worked, too.
- jeroenhd 3y agoI think the problem is that AI is wrong so often that it would be foolish for it to stick to its guns when you're correcting it. Bing Chat did this at first, and started calling the user a manipulator and an abuser when you corrected it too much (regardless of whether the previous message made any sense or not). I found it really funny, but other people were distressed by it because I suppose they thought they were talking to something sentient.
- kgeist 3y agoThe problem is probably specific to the default system prompt in the default ChatGPT UI version. An open weights LLM can be made into a pretty engaging bot with the right prompt/model/finetuning. I use an uncensored LLM with a special system prompt which makes it answer directly without all the fluff.