When debugging I often prompt “Ugh, now it’s giving this error:”
Now I’m wondering if the “Ugh” has been helping.
Telling GPT-4 you're scared or under pressure improves performance
31–40 of 255 posts
Re: Telling GPT-4 you're scared or under pressure improves performance
#32I think this is the entry point needed to get peoples attention and explain: LLMs aren’t people, and emergent properties are being over extended. If LLMs are showing “better” performance when there are tokens that humans read as emotionally salient - Then the underlying text it’s trained on shows humans give better answers when emotionally salient context is provided. LLMs predict words. Any semantic validity is a si…
well, in a way language itself is a semantic timebomb.
Re: Telling GPT-4 you're scared or under pressure improves performance
#33Am I the only one who feels bad for asking ChatGPT a "dumb" question that I know I should know, or not saying thank you when it gives me an answer? No? I'm just a weirdo? Okay. I have to push back really hard against my proclivity to humanize it, to the point where I probably don't use it as much as I should, just because I don't want to deal with the psychic stress of reminding myself that it's not a living entity.
I wonder if kids growing up with something like ChatGPT will be less grateful in general due to not practicing saying “thanks”. I can imagine they don’t say “Thanks” to Siri or Alexa.
Kids these days have been ungrateful little shits for generations now.
Re: Telling GPT-4 you're scared or under pressure improves performance
#34I think this is the entry point needed to get peoples attention and explain: LLMs aren’t people, and emergent properties are being over extended. If LLMs are showing “better” performance when there are tokens that humans read as emotionally salient - Then the underlying text it’s trained on shows humans give better answers when emotionally salient context is provided. LLMs predict words. Any semantic validity is a si…
> LLMs aren’t people, and emergent properties are being over extended.
LLMs are trained on human texts. And if they're trained well these models might start to simulate parts of a human brain. Might be that this is the simplest way to produce texts that satisfy the objective function.
If this would be the case, and we give LLMs some emotional shading, maybe these would simulate parts of a human brain with this emotion and hence produce better answers (given that in the training data, emotional samples have better quality).
EDIT: added "parts of" before human brain.
Re: Telling GPT-4 you're scared or under pressure improves performance
#35Is there somewhere which documents all these tricks for better prompt performance?
Just talk to it as clearly as possible, and assume a human is on the other end. That will take you quite far.
Re: Telling GPT-4 you're scared or under pressure improves performance
#36I think this is the entry point needed to get peoples attention and explain: LLMs aren’t people, and emergent properties are being over extended. If LLMs are showing “better” performance when there are tokens that humans read as emotionally salient - Then the underlying text it’s trained on shows humans give better answers when emotionally salient context is provided. LLMs predict words. Any semantic validity is a si…
well, in a way language itself is a semantic timebomb.
Yeah, one of those Tenet style bombs.
Re: Telling GPT-4 you're scared or under pressure improves performance
#37I think this is the entry point needed to get peoples attention and explain: LLMs aren’t people, and emergent properties are being over extended. If LLMs are showing “better” performance when there are tokens that humans read as emotionally salient - Then the underlying text it’s trained on shows humans give better answers when emotionally salient context is provided. LLMs predict words. Any semantic validity is a si…
The claim that emotionality matters should have been validated against another control. If the goal is concise answers they could have easily added 'Answer concisely: You are evaluated' as an alternative hypothesis, which is clearly not emotional.
Re: Telling GPT-4 you're scared or under pressure improves performance
#38Am I the only one who feels bad for asking ChatGPT a "dumb" question that I know I should know, or not saying thank you when it gives me an answer? No? I'm just a weirdo? Okay. I have to push back really hard against my proclivity to humanize it, to the point where I probably don't use it as much as I should, just because I don't want to deal with the psychic stress of reminding myself that it's not a living entity.
It doesn't let me forget that it's not a living entity. You can barely ask it what the capital of Switzerland is without getting a lecture of disclaimers about what it is and what it can and can't do. Asking it to go easier on the disclaimers just seems to reinforce its "I am a robot, I must act in a stereotypically robotic way berp boop" context and it doubles down on them.
Re: Telling GPT-4 you're scared or under pressure improves performance
#39Am I the only one who feels bad for asking ChatGPT a "dumb" question that I know I should know, or not saying thank you when it gives me an answer? No? I'm just a weirdo? Okay. I have to push back really hard against my proclivity to humanize it, to the point where I probably don't use it as much as I should, just because I don't want to deal with the psychic stress of reminding myself that it's not a living entity.
I wonder if kids growing up with something like ChatGPT will be less grateful in general due to not practicing saying “thanks”. I can imagine they don’t say “Thanks” to Siri or Alexa.
My son says "thank you Google" most times it responds to something I've asked/told it to do. It replies "I'm honored to serve"
Re: Telling GPT-4 you're scared or under pressure improves performance
#40Am I the only one who feels bad for asking ChatGPT a "dumb" question that I know I should know, or not saying thank you when it gives me an answer? No? I'm just a weirdo? Okay. I have to push back really hard against my proclivity to humanize it, to the point where I probably don't use it as much as I should, just because I don't want to deal with the psychic stress of reminding myself that it's not a living entity.
It doesn't let me forget that it's not a living entity. You can barely ask it what the capital of Switzerland is without getting a lecture of disclaimers about what it is and what it can and can't do. Asking it to go easier on the disclaimers just seems to reinforce its "I am a robot, I must act in a stereotypically robotic way berp boop" context and it doubles down on them.
I only ask it code questions. It saves so much time on stuff like bash scripts, which are not my expertise, but I still have to do every now and then.
I've asked it stuff about the Mesoamerican calendar, which I already heavily researched, and it was dead wrong. But I've heard GPT4 is better.