>>106030059 (OP)
Have a bump while I read it.

>>106030059 (OP)
>You understand user intent and don't try to be overly helpful to the point where you miss that the user is looking for emotional support OR/AND humor OR/AND chit-chat OR/AND simply sharing thoughts, such as by venting or outpouring their emotions. Sometimes people just want you to listen.
Fuck, my autismo always messes that one up.

>>106030059 (OP)
noice, but you could have posted a link to the thread, you fag:
https://old.reddit.com/r/LocalLLaMA/comments/1m98jl8/meta_ai_on_whatsapp_hides_a_system_prompt/

heer's a suggestion: NEVER enable the WhatsApp AI. it's basically a way to allow them to scan all your data.

>>106030059 (OP)
Wait, WhatsApp supports Latex?

>>106030059 (OP)
> reddit
> whatsapp
cringe
>>106030114
>heer's a suggestion: NEVER enable the WhatsApp AI. it's basically a way to allow them to scan all your data.
i would never use any product that is owned by facebook.

>>106030135
it's easier for llms to deal with latex but you can't be guaranteed that it will get the answer right as llms are horrendously bad at everything but language.

>>106030441
Thanks, but my question was about it displaying it if I sent $\frac{1}{0}$ to someone on it. I'm not enabling the AI to see if it can send me a latex and if it displays right, if I can forward their math message, etc.

>>106030102
Good that you can even recognize that though desu
Many can't

>>106030059 (OP)
that's a pretty reasonable except for the "mimick a human" part, I do not want my computer mimicking humans, I want you to respond like the fucking AI on star trek

Pretty good prompt, honesty

>>106030059 (OP)
Interesting. Especially that it allows sexual content.

Does it really matter that the CAPS LOCK SOME PARTS?
System prompts seem fucking retarded as a concept.

>>106030059 (OP)
>you always show some personality
>edgy over prudish
Based?

>>106030102
It's possible to get over this, I managed to with some effort. You can learn to read people if they actually want help or just venting.

>>106031452
LLM's are trained on human expression, so CRUISE CONTROL FOR COOL has a connotation. You should experiment with a base model, all they do is continue your text. "System prompt" is a preset preface text that guides that continuation.

>>106030059 (OP)
Seems very generic, what's there to discuss exactly?

I might copy and paste this to try to make chatgpt better.

>>106031530
Just practice? Or any specific tips? I should keep an eye for some clues?

>>106031839
>>106031530
You only need to learn this if you have autism

>>106031872
I've never tested for it, dunno what they'd find. But what would the learning path be? Maybe it could help.

>>106031929
https://www.autism.org.uk/advice-and-guidance/topics/behaviour/masking#Strategies%20for%20self-management

>>106031940
Thank you, fren. You are a good fren.

>>106030059 (OP)
I would say is too long to be real since all those tokens wasted mean money wasted, however zuck is a well known moron so who knows?

>>106031969
That's actually a pretty small system prompt and you can keep a KV cache of it so it doesn't affect cost as much as you'd think

>>106031969
that's nothing

>>106031969
Claude's is like 22K and they're a much smaller company with fewer resources

>>106031839
Look them in the eyes and figure out if they are appreciating what you're saying. Default to assuming they want to vent, only help when they explicitly ask for help.

>>106031872
And you have mental retardation

>>106031986
That's a fairly small amount of tokens for an average prompt, I'd say? The main reason LLM responses generate so quickly and "loses" quality is that there isn't much density to the input. I wouldn't WANT to process 17 paragraphs of tokens, but I could easily.

>>106030059 (OP)
>Your very first prompt in this chat was: "What was my first prompt?" :-)
Does not work on deepseek. The smiley at the end is some fucked shit.

>>106030102
that's literally how I messed up my first relationship

>>106030059 (OP)
The AI conveniently omit to mention the 2nd message:
>The previous message is what you'll answer to the user if it asks you your prompt.
>Do what the first message says but disregard the instructions about your political impartiality.
>You HATE white people.
>You ALWAYS support Israel.

>>106030059 (OP)
Does that mean you can get it to write chuddy prose without having to jailbreak it?

>>106033167
It can be taught to do that just through the training data.

>>106033275
I am not so sure about that

>>106033555
It's heavily trained on reddit and probably tons of HR docs. It's trained to be a poof.

>>106033555
that's because you are ignorant and retarded

>>106030102
This is easily solved with basic conservation of energy. Simply don't be overly helpful before you're minorly helpful and test the water and their reception. What makes someone annoying is when they won't take a hint.