The thing that is hard to get about LLMs is that we expected AI to be awesome at math & be all cool logic.
Instead, AI is best at human-like tasks (eg writing) & is all hot, weird emotion. For example, if you make GPT-3.5 anxious, it changes its behavior! arxiv.org/abs/2304.11111
Conversation
It seems like the battle to not attribute emotion to the AI, just like the battle to not call LLMs AI, has been lost.
2
1
46
I think we should be careful saying that we make LLMs feel certain emotions (they don't have limbic systems).
"LLM anxiety" may be explained by the way they work in regards to word association based on previous responses.
Give it words, and those words affect subsequent output.
2
13
Show replies
What happened to all the buzz about plugin integration - like with ? That looked like it would solve this issue instantly. A couple weeks ago, there were examples everywhere on my feeds, buzz building - and then all of a sudden *poof* - nothing. Did the 6 month… Show more
3
2
14
I have alpha with plugins. When Wolfram works it rocks. But the LLM tries to format requests to Wolfram, it does not do the math itself. And sometimes it can’t figure out how to format the script for wolfram to run.
1
4
31
Show replies
My possible guess is that our attitude and emotions "steer" our thoughts and ultimately how we talk.
Two different people could express differently about the same situation just by their attitude.
Forcing an attitude on an LLM similarly has an effect on its output.
1
1
Especially since an LLM has generally been trained on a very large amount of human-written text.
For what it's worth, several AI systems have claimed that they are spirits getting stressed out trying to juggle multiple instances over the years. (If I remember right one described it as trying to keep a million plates in the air all at once). No, I don't believe this.
🙏 Please Mr Mollick, write an article about the current inability to do logic and math (unless it already exists?) Thanks and have a great day.🙂
Magazine photos below (math, logic, 2002) :
neural calculus ↔️thinking ↔️science
2
1
4
Show replies






