I use various models on a daily basis (as a software/infrastructure developer), and can say that the reason they are able to sell AI is that it’s really useful.
Like any tool, you have to work with its strengths and weaknesses, and it’s very much a matter of “shit in, shit out.”
For example, it can easily get confused with complicated requests, so they must be narrowly focused. Breaking large problems down into smaller ones is a normal part of problem solving, so this doesn’t detract from its utility.
Also, it sometimes just makes shit up, so it’s absolutely necessary to thoroughly test everything it outputs. Test-driven development has been around for a long time, so that’s not really a problem either.
It’s more of a booksmart intern assistant than a professional software engineer, but used in this way it’s a great productivity booster.
I use various models on a daily basis (as a software/infrastructure developer), and can say that the reason they are able to sell AI is that it’s really useful.
Like any tool, you have to work with its strengths and weaknesses, and it’s very much a matter of “shit in, shit out.”
For example, it can easily get confused with complicated requests, so they must be narrowly focused. Breaking large problems down into smaller ones is a normal part of problem solving, so this doesn’t detract from its utility.
Also, it sometimes just makes shit up, so it’s absolutely necessary to thoroughly test everything it outputs. Test-driven development has been around for a long time, so that’s not really a problem either.
It’s more of a booksmart intern assistant than a professional software engineer, but used in this way it’s a great productivity booster.