You could start by testing it.
I already have:
Some C libraries that I know well, AI can't generate code for at all.
It gets names of functions wrong and replaces them with them that sound good, but are not. But it is not functional code at all. Now, if I was probably doing a python program, it probably has plenty to scrap and create something at least somewhat usable. However, again given post scrapping cleanup, is it still worth it? This is perhaps the biggest problem with even the current auto conversion tools that most on here are used to.
And here as well:
Now, given the group here, if talking about art, it does horrid things there, especially with the extremities, but even things like zippers "melting" into skin or just how some body parts attach to other body parts.
I didn't pull this out of thin air.
Yes, I have tested it. If nothing else to be able to cogently try to form an opinion on it.
Now, I'm sure that with any tech, there will be advancements and what I have tested here may not be applicable in a few months time, sure, but it's still doing it now. And while they improve the LLMs, there will be people that will try to poison the well. I can see some things being behind more paywalls and not so easily accessible as it once was or it will just stop being done at all and that would be less for the LLMs to "learn" from.
The biggest thing would be in house version and that would probably be a good middle ground. Don't know if that would happen or not. Sure may have open source ones that could be done that way, I think llama is one. I have a feeling that in some areas, if the laws don't change, that will probably be the route that most go, is in house versions.
I bet that's what everyone said about the internet & the knowledge base vs the current libraries of encyclopedias,
The closest analogy would be that of doing digital art versus traditional pen and paper. The difference between that point and now, is a whole lot of abstraction, much more compared to what it was just with going from traditional to digital. This isn't even going into the legal issues (at least here stateside, although I'm sure that will change and this would apply for code as well as art). The more the user knows what is being abstracted, the better the end result with something like LLMs and they of course, will be able to clean that up in post. The less one knows, it's going to be a rough time.
The problem is that the people that have the best chance of that are the users now, the 3rd and 4th gen users, actually not so much. We have seen that with even computer usage with people that are what have been called "digital natives". Not everyone, but an astounding percentage of them are that way. They aren't even used to a file system and how to interact with it. That would be more of my concern happening with this tool. Just like we have people that can really only use auto conversion and don't know how to clean things up after that conversion.
The one problem that I do have (and it has been illustrated in a bad way in recent years) is that it's too easy and quickly to change things to suit the now with regard to your quoted analogy. Some of that can be good, but it does depend on the application. So always having that constant ability to fiddle with things that is "live" like on the internet, can be bad. Not all instances, but enough of them have happened that make really need to vet sources that much more (a little bit harder to do if one doesn't have the knowledge base to vet them though, especially if one is learning something).
Wild West is keeping us up on new tech and opinions.
There will be a test on Friday.
No, there won't be a test. I just think that it's good to keep up with what the makers of tools that we depend on and in what direction that they are going in. Some may agree with, some may not. That is all.
