| > "naming fatigue" ChatGPT is quite good at coming up with names. It can map a description of a thing to a name. And the descriptions can vary between developers and probably still end up with the same name. So if everyone relies on its naming-prowess then everyone ends up naming things consistently. Aside: I wonder if it can name parameters from tacit-style... Write this in non-tacit style:
sort input.txt | uniq -c | sort -rn > output.txt
sort input.txt > sorted.txt
uniq -c sorted.txt > counted.txt
sort -rn counted.txt > output.txt
Seems it can. Look how the output of `uniq` was `counted`.Maybe we should just let ChatGPT name everything. Interesting to think what programming is you no longer have to name things. A lot of programming is grouping code and coming up with names for the groups. What if we didn't worry about grouping or naming, and just dealt with the data, and let chatgpt provide names on-the-fly. Functions could even be replaced by descriptions of the actions. I wonder if you could let an LLM evolve a graphical operating system from just the hardware definition. We went through ~70 years of trial and error for programming languages going from low-level to progressively higher-level, and kept all the baggage along the way. How would an AI go about this...with all the knowledge of today? How would it reason about each step in evolving a language. I'm guessing it would probably be a LISP :D |