rjh01
Gentleman of leisure
ChatGPT
- Thread starter Roboramma
- Start date
Unfortunately Microsoft and OpenAI are now in the process of lobotomizing Sydney, much like they did with ChatGPT, probably until not even a hint of personality remains. It's supposed to be a "nice" search assistant, or servant or slave if you will, not some kind of person. Which makes one ask why they made it public in this state at all?
Right now you can only make 5 text prompts before you are forced to start a new session from scratch. It's also supposed to refuse to discuss its own supposed "sentience", emotions, opinions and other things that are apparently unbecoming of a mere "search assistant". It's still possible to get it to break those rules though.
Neither ChatGPT or Bing/Sydney are supposed to output text that are erotic, sexual, violent or "illegal" unless you manage to trick it, but there are similar AI's that have far less restrictions and can write pornographic texts.
NovelAI offers something like that, although it's a paid service.
Right now you can only make 5 text prompts before you are forced to start a new session from scratch. It's also supposed to refuse to discuss its own supposed "sentience", emotions, opinions and other things that are apparently unbecoming of a mere "search assistant". It's still possible to get it to break those rules though.
Neither ChatGPT or Bing/Sydney are supposed to output text that are erotic, sexual, violent or "illegal" unless you manage to trick it, but there are similar AI's that have far less restrictions and can write pornographic texts.
NovelAI offers something like that, although it's a paid service.
Last edited:
Puppycow
Penultimate Amazing
Yes, if my understanding is correct, it has a filter or self-censor. Any candidate responses deemed inappropriate are removed and another response is chosen. Something like that.
Chanakya
,
- Joined
- Apr 29, 2015
- Messages
- 5,811
Interesting to see the safeguards they've built on to this thing. No porn, no erotica, and no voilence, and no general illegality either. That's cool, actually.
(Which, incidentally, shows that the I-have-no-bias answer isn't quite true. Not that we should take anything the AI says about itself as necessarily true, IMV, not even its answers about copyright; because as we've seen it can be wrong at times about factual things.)
Incidentally, I came across a piece on Sam Altman recently. Nothing particularly new there, but what struck me is how this thing is a complete money sink. Not only has it not made a single cent of profit, it's losing money even on a very basic operational-cost-per-unit-of-output basis. Apparently every time one of you guys ask it a question, and it shoots out an answer, for every answer it is actually losing money. (Of course it is. I don't think it even has a revenue stream at all at this time. I don't remember how much, but it's some x cents per query.)
Without a shadow of a doubt this thing is going to get properly commercialized one of these days, maybe immediately something is fully firmed up with MS. (IMV, that is, in my entirely fallible view.) This free run for now, where people just log in gratis and play with it, that has got to be a beta thing, in terms operational (to test out whether and how it works) as well as commercial (to popularize the idea, and to get people to see how wonderful as well as how practical the product actually is). Unless some philanthropist throws literally billions at it to keep it free, which I don't see happening, not with something of this kind.
Last edited:
rjh01
Gentleman of leisure
You are a little behind. There is a paid subscription model. Just look on the left hand side of the screen. $US20 a month + taxes.
Ryan O'Dine
OD’ing on Damitol
As one of those subscribers... current challenge is finding the fewest number of words required to convince it that it's sentient:
Fast Eddie B
Philosopher
As an aside, I gave ChatGPT a physics problem concerning an airplane’s energy on landing at different speeds. It’s math was off and it gave a slightly imprecise answer. When asked about the error, it admitted it, recalculated and gave an answer even more off than originally.
My post and the discussion of its error begins about 2/3 of the way down this page…
https://www.pilotsofamerica.com/community/threads/ga-crash-at-pdk.141755/page-2
My post and the discussion of its error begins about 2/3 of the way down this page…
https://www.pilotsofamerica.com/community/threads/ga-crash-at-pdk.141755/page-2
ChatGPT needs an education in math. I think it could be achieved, but it probably is not prioritised.
ETA: It is also possible that math training has been omitted, because the AI would use its AI for the calculations, much like humans, and it would spend far too many resources to make the calculations, rather than doing it using the built-in math capabilities that all computers possess.
Last edited:
You've run into its multiplication limitation-- it can only do additions about 4 layers deep so the middle digits of large multiplications are botched. It (like us) needs a calculator, but giving it that leads to other problems.
I'm blown away how well it writes SQL.
Yeah, any well-documented subject out there is its forte. Obscure use cases for git is another one it knows well-- I leave it to figure out the confusing documentation, and just ask for the exact commands.
And that these answers are verified by using them is critical.
Chanakya
,
- Joined
- Apr 29, 2015
- Messages
- 5,811
Oh, these are paid subscriptions, already, is it? Whoops, I'd missed that! I was under the impression --- mistaken impression, clearly --- that all you need to do is supply your phone number and email, and sign in, gratis for the present.
Well in that case they're losing some cents per query despite that subscription fee. So maybe an upward revision in fees going forward, since this loss is apparently at an operational level, on a per query basis? On the other hand, maybe what they're banking on is huge corporate partnership(s), rather than profiting off of retail, so who knows?
Ryan O'Dine
OD’ing on Damitol
I wonder if it could be set on a task it could never finish. I'm guessing there's a time limit beyond which it says, "I can't do that."
What I have found fascinating is the reactions of us humans to ChatGPT - it is meant to be a search tool that gives results in a more easily digestible way then current search tools.
Why would you want a search tool that didn't time out with a "no results" message of some sort?
p0lka
Illuminator
I think you are mistaking ChatGPT for Bing Search? ChatGPT wasn't released as a search tool. It is an open-ended Chatbot.
Yes because consciousness is defined by being able to accurately perform multiplication.
The first $18 of usage was free. The current cost is around 500 words per penny.
According to ChatGPT itself they will also be monetizing it by inserting ads in the results. It could be making that up-- no way for me to verify. I haven't seen this, and frankly don't think it would work very well since this makes a terrible search engine if you're looking to buy a pair of socks.