No kidding. Chatgpt couldn't even replace the last couple pages within 42 word docs that I asked it to, using content in another word doc. Copilot wasn't much better. No matter how many times I prompted both it found a new way to fuck the formatting every time. Even after explicitly telling to keep the formatting exactly the same.with the amount of mistakes chatgpt makes........... you're only slightly better than doing it yourself.... even the highly touted Grok4 produces mistakes, and a small mistake can turn into a monster, big mistakes can be catastrophic if the scope of the project isnt completely understood.
That sounds like the best type of use for it.The reason I use it is because it can make simple and clean code, for instance, I can give it to produce a piece of code that calls the database based on some parameters, lock for given id, update sql, move to the next, and include logger object for logging, and it does that framework very well, with notes, declarations, etc, and I may end up saving time on those repetitive operations.
The difference between the AI and a human, is that the AI isnt going to tell me when I have a problem that should be solved with well packaged data vs a problem that should be solved with software coding, and that's similar to entry level developers, they'll omit important factors of architecting the solution.
The problem is... When you feed too much shit to it over time... It causes the model to collapse. It kicks out whatever you ask, right or wrong and it hallucinates even more. It's like to gotta keep it in line so it doesn't spit out nonsenseThat sounds like the best type of use for it.
I gotta wonder, if all these stories of mediocrity like from @TeTe above is intentionally built in, in order to not overwhelm folks and have them remain comfortable enough with it to keep using it and, constantly, keep training it.
I bet the infrastructure running them can barely keep up with demand/usage. The best case would be to figure out how to run it offline on a local machine so a user wouldn't be exposed to whatever issues arise from remote access along with everyone else.The problem is... When you feed too much shit to it over time... It causes the model to collapse. It kicks out whatever you ask, right or wrong and it hallucinates even more. It's like to gotta keep it in line so it doesn't spit out nonsense
Oh and anyone else who wants to purchase that data or whatever else happens to that stuffI bet the infrastructure running them can barely keep up with demand/usage. The best case would be to figure out how to run it offline on a local machine so a user wouldn't be exposed to whatever issues arise from remote access along with everyone else.
It'd also remove the creep factor of all your interactions logged forever by companies A-F making it, and their marketing partners G-ZZ as well as world governements 1-130
A recruiter from a big company asked if I would like to be considered for a role. A buddy of mine is working for this company, it will be a public company sometime in the future so I can join and gain pre-IPO equity which would be pretty sweet. But knowing myself, I'll probably last a year before I give myself some reasons on why I should quit. Maybe I can ask in the interview if I can have an output based job where I don't necessarily need to clock in and out everyday, but I doubt it.