The big LLMs are expensive because they work well, and to work well they have to use a lot of resources. I haven't used any local LLMs but I don't really see the point since the big ones work so well. It seems to me going local is just going to invite lots of problems and to eliminate those problems you will just have to end up spending more on them anyway. There doesn't seem to be any way around the old adage "you get what you pay for".

You are taking about GitHub copilot in vscode? I never need to add files, I just ask it questions and it will search through all the files if it has to.

What a great day for rich bullies everywhere!

I think most of their parents were probably like Trump, so they grew up to see that kind of behavior as protection.

Are they shaped like mushrooms?

The base HATES lies. But the problem is they almost always get truth and lies backwards.

Studies show that human IQ has fallen considerably over the last decade. I'm afraid nature and/or the chemicals in the air that create dumber babies isn't on your side.

I didn't ask whether it took skill. I know it takes skill. .I asked why AI is considered a skill.

Well if they now need to mind the quality than that's good all around.

That doesn't really require people skills. You can be an absolute robot with no personality and still ask the right questions and write down the right answers... They don't have to like you.

If I was looking for a job now and had to write a resume, I would either get a popular AI to score it or just get the AI to rewrite it so that it scores highly. Most of them would pick similar in that regard.

Here is what I have observed; there are people that will design a browser (possibly also using AI) and then walk the AI through all the steps. Then there are people who just ask it to build a browser. The former may pay $500 for the AI and the latter will pay $10,000.

I used Microsoft Copilot at first and it was alright. It only takes three files to reference at a time but you can tell it that you are uploading more and to wait. That was already much better for small projects, and I could use it for my large project but I had to be careful to upload all the scripts in the paths it needed to figure out and that was a bit onourus. I get Github Copilot through work though and that takes away all the manual work to upload the files wihich makes me a lot faster. Also it is better at accessing the terminal and analysing external websites on its own too.

Well yeah, if you are a manual ditch digger and someone invents the excavator then you have to buy an excavator. Love or hate machinery, that's the way technology has always been.

Yet I have found AI to be incredibly powerful. I just finished a ~17k line application and the last 4000 lines or so would have taken me a few months but AI did it in a few weeks. Then I made a website for it. found a Wordpress template I liked but I didn't want Wordpress so I asked AI to tear out the Wordpress console and just make it a static site. AI traced my code and wrote all the documentation for me as it could give explanations for the fields by referencing documentation for the end result. Yes I'm sure the HTML is nasty but I never had to look at it because I just built the whole site with prompts. And yes, sure you have to test but no differently than I would have tested my own code anyway. The most frustrating thing is that it would occasionally regress changes that I had already added but I would tell it to look back to where I added it the first time and it would be fixed. As for having to specify everything, generally you can keep one file of specifications and it can reference that.. To me that is just a communication skill.