The fun part is when it hallucinates a library like that and it's simple enough that you can ask it to implement it from its own usage examples. I got that working with a Sequel (Ruby ORM) plugin that it insisted existed that didn't - the implementation I asked it for only needed a couple of minor tweaks.
I got it to write a Python module that would intercept calls to missing functions, create the missing function based on the name and calling parameters, write the function to a new .py file, modify the file that called the missing function to import the new file, then execute the new function and return without error.
It was very unhappy about doing so; I had to stress that it was just for research in the prompt to avoid it flat out refusing to write self-modifying code. But it worked!
That makes me wonder where you can go if combining GPT with Koza et.al's genetic programming research, to instead of doing "blind" mutation and crossover, let GPT do targeted changes at high temperature, driven by knowing the goal, and pit different runs against each other.
An AI-aided translation might make more sense, since at least you could review that for correctness and it'd likely be easier+cheaper to run the result.
It's no different in everyday life. This was an appropriate title for the Rust thread, but is too terse for Hacker News.
If I say to my mother "Hannah told me she'd be away for Christmas" we both understand who is meant, my sister. If I say that to my friend Dave it's unclear. Which Hannah? We have a mutual friend called Hannah, I have a sister named Hannah, or maybe I mean the Hannah he works with. So I should be more specific in that context.
Being obliged to be fully specific all the time is no problem for a machine but it's tiresome for humans, so no, I do not agree with "death to abbreviations". Maybe HN rules should encourage people to expand abbreviations when citing material from elsewhere that could be unfamiliar to most readers. As it is, hey, free Internet points for whoever first expands the abbreviation in a comment.
I’m on a similar track. Web/java background and started doing some tutorials and aoc, currently working on the the same ray tracer challenge book (it’s awesome). Not sure what a next step would be, maybe that one book where you implement a web browser or some raspberry pi project. Any recommendations?
I'm fairly certain HN has almost enough people that they could make it a viable venue (geography-contextually, naturally). Add some meme marketing to it with some vague overtones about making the world love something they didn't love before and it'll get memorialized as a TEDtalk.