Technology

It took a while but yeah that seems about right. It takes a lot of guiding to have it produce something usable. I have to know a lot about what I want it to do. It can teach me things but the hallucinations are strong sometimes so you have to be careful.

Still it helps me out and I make a lot of progress because of it.

That very much depends on if you believed all the hype or not. If you did, then yes, it's failing, as ChatGPT was supposed to be the next big breakthrough that was going to automate everything ever, and any company that didn't get in on that right now was going to be left in the dust by all their competitors. On the other hand, if you were an actual sane person (so you know, not a CTO/CEO), then this is very much a non-story as you always knew that all those outlandish claims were nonsense and that this was always going to be yet another niche piece of tech that's useful in a few places in limited amounts.

FWIW, it’s become an inextricable part of my life. I use it for hours every day, for programming and Linux advice, spreadsheet help, foreign language practice, and random trivia.

Yesterday, I discovered that Snoop Dogg’s -izz speak from the aughts was actually derived from carny pig Latin.

Nah, that sounds about right. This is just the natural result of people actually trying to use GPT for all the things they were told it would be able to do, and now discovering that was in fact all bullshit. The LLMs are and always have been massively overhyped and oversold on what they can do. Sadly this won't stop the corporate executives from trying to use them to replace workers, although when that effort eventually face plants they'll just quietly re-hire a bunch of people and find some middle manager to blame for their failure. This is was and always will be merely a productivity tool to automate some repetitive work, but it still needs someone to review and clean up its output. It's not "replace someone doing 40 hours of work a week", it's "allow someone to do what used to take 40 hours in 35 hours instead".

Sadly the most impact this is going to have is on spammers and scammers, who can now automate generating their garbage since it never mattered that any of that crap was accurate or not, merely that on a casual glance it looks reasonable.

When I first started messing with it, I was kind of neat and fun. I like making characters so I was using it for like story prompts and general outlines. Some were better than others, but it was neat for some inspiration and fleshing out. I never took it's outputs 1:1.

But when I messed with it again recently. It was a lot worse. Like it ignored parts of my prompt. Like as an example a prompt was about a romance story, but the story was about character A and their family. The love interest character was barely a footnote and could have been removed entirely and nothing would have changed with the story outlines it was giving me.

I thought maybe it doesnt like romance prompts, so I tried less specific and more broad prompts from there, and it was the same thing of just... not outputting what I was asking it to. It got worse and worse and sometimes wouldn't output anything at all.

Not sure what language you're coding in, but I've found GPT-4 incredibly helpful for coding in C++ and Python.

There's also the novelty factor. Like how DALL-E was the rage not that long ago, people flocked to it because it was new and interesting.

But the novelty has rather worn out by now.

It is so neutered now that I rarely find it is worth the time to use it. Now it just gives non-answers that are rarely helpful or accurate.

Edit: To give a bit more context, I still use it several times a day just to be sure it is still a disappointment.

It was always able to do some genuinely amazing things, but it was always limited when you took it beyond its wheelhouse. It helps to think of it not as an "AI" as people keep saying, but as a "text completer" with a huge amount of power within that domain.

Or, another way to think of it is as a super-powerful search engine. If the answers and knowledge you're asking of it were fed into it as input data at some point in its training, it'll probably be able to find it and reformat it back to you with a scary amount of smoothness and precision. If you're asking it to figure out something new, it may be able to fake it in some short-term fashion or another based on what it's seen, but not with any genuine understanding behind it. That's just not what it does. I actually have a little private theory that if it was given something like the bar exam in scale and complexity, but an exam was genuinely a whole new novel invention that hadn't been extensively discussed and represented in its input corpus, it would fail pretty badly. A lot of what humans can do that makes them capable is adapt to new domains -- we can teach ourselves to play chess, or do math, or fly airplanes, or play Celeste. GPT is hugely impressive but it's still only one domain.

I actually don't believe that it's gotten substantively less capable. I think there are little ticks up and down in its capability sometimes in particular areas, and people seize on those to conclude that it's now becoming dumber, but in my experience, the raw API was always quite capable (more so than the somewhat nerfed chat interface), and it was always super-capable with some tasks and not at all capable with others. I think journalists are just now figuring out that, after having studied the issue in their professional capacity for the better part of a year, and reporting on it as if it's a new thing.

From what I've seen, here's what happened. GPT 4 came out, and it can pass the bar exam and medical boards. Then more recently some studies came out. Some of them from before GPT 4 was released that just finally got out or picked up by the press, others that were poorly done or used GPT 3 (probably because of gpt 4 being expensive) and the press doesn't pick up on the difference. Gpt 4 is really good and has lots of uses. Gpt 3 has many uses as well but is definitely way more prone to hallucinating.

Came here to say this. ChatGPT is very good at coming up with convincing bullshit. Always do your own research.

Just wait until we find out what the US military has.

Cheating isn’t necessarily the only use case for GPT, although it definitely does play an impact on the overall number of users.

He looks like the Change My Mind meme guy.

Do you know of any interfaces like (or better than) ChatGPT that you can selfhost, that uses the openai api?

Yeah, that's what I did. With my very light usage the fixed-price subscription isn't justifiable, but the api works nicely.