I've been repeating again and again that everywhere, but the sweet spot for beginners is to stick to the boostrapping method that will lead to the least possible modes of failure.
Because I don't want to repeat it every time, I eventually wrote everything down, so know I can just link to it:
For installing Python: "Installing and running Python: the bare minimum you can get away with"
At this point it kinda feels like spamming, but honestly, there is too much noise out there, and those points don't come across enough. Especially since most dev are either too little or too much experienced to be helpful.
We tried the alternative, and it's less productive.
At some point, there is the theory and practice.
Since LLM output are anything but an exact science from the users perspective, trials and errors are what's up.
You can state all day long how it works internally and how people should use it, but people I've not waited for you, they used it intensively, for million of hours.
I have my own sense of morality developed over years of balancing life, I don't want a robot to remind me of the average moral construct present in its training data. It's noise.
Just like I don't want a hammer to keep reminding me I could hit my fingers.
I'm pretty sure the moral stuff is there because a lot of previous chatbots were tricked to make questionable responses that led to bad press. So OpenAI put a lot of effort into making their models behave ethical.
I'm not sure why people are so annoyed by the ethical disclaimers. I don't even recall running into them, but maybe that's just because I don't ask controversial questions.
this whole subthread is an example of things most people would prefer not to see but have successfully trained themselves to ignore over the course of their lives.
excessive moralizing, being reminded that things are there for a reason etc etc
gpt is overly verbose about pointless things, i understand it's by design coz of some person's dogmatic agenda but i don't care personally.
tl;dr - ethics warnings in gpt are like coffee hot warnings on coffee cups
I've tried many system prompts so far, and I'm underwhelmed with the result. Especially, I keep insisting to just give me answers with as little context as possible. E.G, if I ask for code, just give the code.
But the gpt does what the gpt wants.
It's a minor annoyance though, 1st world problem at its best.
Streaming is expensive, but less than you think. If you run your own servers and don't use the cloud, you can serve a million user a month for a few thousand dollars.
This ensures when I drink it, the difference is huge.