If you ask chatgpt for 100 5 letter words around a single topic, instead of saying it can’t give you 100 words it will just start adding 6 letter words and then start cutting off a letter.
It’s astounding how much faith people put in this software
I went to a conference a few months ago and the very first speaker gave the following advice with a straight face to a room full of professional software engineers: “Your biggest limitation on your productivity is going to be token management, so just buy as many tokens as you can so you won’t even have to think about it.” And that guy, supposedly, didn’t work for OpenAI or Anthropic.
I kind of hope he’s at least getting kickbacks because I would rather he be a secret corporate AI shill than just a submissive gimp for dommy mommy AI industry attempting to recruit more paypigs to her flock. At least that would have more dignity.
i always wondered what they are peddling at these AI conferences, we have them almost daily here in the west. im not really surprised they have a hired “spokesman” to do it, are the engineers buying into this? or they know full well the AI isnt shit?
Well, the unfortunate truth is AI is an extremely useful tool for software engineering. I’ve gotten to where I use it most days and it has made some tasks waaaay easier and faster.
But it’s not a silver bullet that solves all your problems and replaces an engineer that understands their projects, business needs, context, inter-team dependencies and agreements, risk mitigation, etc. And we also understand that it will never be cheaper than it is right now and getting too dependent on a tool that may be prohibitively expensive in the future is unwise.
If I were an independent contractor, paid by the job, building a bespoke self contained application for someone where they give me all the context I need for it, I’d 100% be using AI to do the majority of the coding and testing. Get the job done fast and move on. But throwing all of your money at it like it will solve all your problems is just moronic, particularly when you work at an enterprise scale where literally no individual person can give the AI the full context of all our systems.
“Write program worth 1 million dollars. Do not hallucinate. No mistakes. Good code only. Make secure. No vulnerabilities. Follow all standards. No spaghetti code. No anti-patterns. No deprecated dependencies. Runs fast, and cheap, and completely functionally. Does what it is supposed to. Minimize token use.”
Did they include “do not hallucinate” in the prompt? Didnt think so. Classic mistake
Unbelievably, this is real advice I’ve heard from corporate AI experts.
It is included in the guardrails for my orgs copilot integration. Surprisingly, it still hallucinates.
If you ask chatgpt for 100 5 letter words around a single topic, instead of saying it can’t give you 100 words it will just start adding 6 letter words and then start cutting off a letter.
It’s astounding how much faith people put in this software
Well that’s just ridic
That movie was ridic too
Obviously the solution is to tell it to not hallucinate that it isnt hallucinating
it will hallucinate that its not hallucinating.
so tell it to not hallucinate that it isnt hallucinating that it isnt hallucinating.
"Oh, thanks for pointing that out, I 100% hallucinated those connections this last time. Let’s go over what’s real:
“I do not think it means what you think it means.”
That was bad, I.N.T.E.L.L.I.G.E.N.C.E.
I went to a conference a few months ago and the very first speaker gave the following advice with a straight face to a room full of professional software engineers: “Your biggest limitation on your productivity is going to be token management, so just buy as many tokens as you can so you won’t even have to think about it.” And that guy, supposedly, didn’t work for OpenAI or Anthropic.
I kind of hope he’s at least getting kickbacks because I would rather he be a secret corporate AI shill than just a submissive gimp for dommy mommy AI industry attempting to recruit more paypigs to her flock. At least that would have more dignity.
i always wondered what they are peddling at these AI conferences, we have them almost daily here in the west. im not really surprised they have a hired “spokesman” to do it, are the engineers buying into this? or they know full well the AI isnt shit?
Well, the unfortunate truth is AI is an extremely useful tool for software engineering. I’ve gotten to where I use it most days and it has made some tasks waaaay easier and faster.
But it’s not a silver bullet that solves all your problems and replaces an engineer that understands their projects, business needs, context, inter-team dependencies and agreements, risk mitigation, etc. And we also understand that it will never be cheaper than it is right now and getting too dependent on a tool that may be prohibitively expensive in the future is unwise.
If I were an independent contractor, paid by the job, building a bespoke self contained application for someone where they give me all the context I need for it, I’d 100% be using AI to do the majority of the coding and testing. Get the job done fast and move on. But throwing all of your money at it like it will solve all your problems is just moronic, particularly when you work at an enterprise scale where literally no individual person can give the AI the full context of all our systems.
Lulz, amen to that
“Write program worth 1 million dollars. Do not hallucinate. No mistakes. Good code only. Make secure. No vulnerabilities. Follow all standards. No spaghetti code. No anti-patterns. No deprecated dependencies. Runs fast, and cheap, and completely functionally. Does what it is supposed to. Minimize token use.”
Perfect. Iron-clad. Let the profits commence.
This is like not even a parody of what they think lol. It’s so ridiculous.
Got it, this program was worth one million dollars before uber was invented.
the AI would Hallucinate the part where “it doesnt hallucinate”
see my other comment.