It's unclear to me that there is any use for the "hammer" of text generation. It adds no new knowledge, and doesn't pretend to. Its transformations of existing knowledge are neither interesting nor attractive. Anything they say, has already been better said elsewhere.
I can imagine uses for generated art, which may at least be aesthetically pleasing. But I can't conceive of any end for computer generated text.
Its use is very much in question - but it is certainly a powerful tool, and that combination is worrisome. Much like the social graph, it is going to have a profound impact on how we interact online and with each other, and that impact will not be known for some time, even though we may be feeling it now, unawares. In a decade, maybe less, we will have some picture of the use and power of these models, there will be meetings in front of congress on how tech companies have used them, etc.
Just look at what is happening in education right now. It is ultimately going to force a complete reinvention of the written assignment. This is just the beginning, even if the tool appears to be a mostly-useless toy for any real-world applications.
The fact that it has a notable downside is certainly interesting. School assignments are well suited to LLMs because they also don't present new information. That's not what they're for; they're for assessing what the student knows.
They're usually fairly obvious, but it's hard to prove. Unlike much ordinary copypaste plagiarism, you can't trivially reject it as cheating. That forces teachers to think of new ways to test student knowledge... an interesting challenge, if not exactly a "use".
> It is ultimately going to force a complete reinvention of the written assignment.
If by complete reinvention you mean returning to what we used to do, which is write essays with a pencil during class without using a computer.
> This is just the beginning, even if the tool appears to be a mostly-useless toy for any real-world applications.
It is not possible to tell on this side where LLMs (or any invention) fall on the spectrum of 3D tv to the smartphone. It will become apparent in the future and 50% of us will have been wrong but anyone who claims to know is just BSing.
I started to do some cpp development again and google is just giving me wrong/outdated solutions or just unanswered question from SO, while ChatGPT is on point most of the times.
Personally, I think that the beauty of current LLMs are their ability to process and present information. Although current generation LLMs might not be best suited to making new discoveries or producing new (valuable) information, their ability to summarize and process information already out in the open is undeniably valuable.
It can do a lot of cool things! You can build a house with it, you can smith metal with it, and you can even use it as a weapon.
The thing is, right now, we're so amazed by its potential that we're finding a lot of uses that, while technically possible, aren't a great fit.
Technically you can use the hammer as an axe, a hole digger, and a backscratcher, but there are far better tools for the job.