ToeMissile
Pronouns: zie/zhem/zer
- 3,169
- 2,058
I assume you can use verbiage more in line with "highly/poorly rated" vs good/bad
- 1
Chat GPT AI
- Thread starter Sanrith Descartes
- Start date
-
Guest, it's time once again for the massively important and exciting FoH Asshat Tournament!
Go here and give us your nominations!Who's been the biggest Asshat in the last year? Give us your worst ones!
ShakyJake
<Donor>
- 7,912
- 19,957
Edaw
Parody
- 13,272
- 88,000
Confirms what I said earlier. It only has links for what is in it's dataset. So no revised data.
- 1
Mist
REEEEeyore
- 31,198
- 23,369
ShakyJake
<Donor>
- 7,912
- 19,957
I have had it inspect data from Pastebin and GitHub so it can indeed follow links. In fact, I did it just a little bit ago, but it took some coaxing. Granted, those two sites have an API interface so maybe that's the difference? Regardless, it gets in a state of confusion of what it can and cannot do.
- 1
pharmakos
soʞɐɯɹɐɥd
- 16,305
- -2,234
If the links are more than 2 years old then they're possibly in its training data, is what we're saying.
ShakyJake
<Donor>
- 7,912
- 19,957
Okay, fine. But the links I've given it were ones I created within the past couple days.
- 1
pharmakos
soʞɐɯɹɐɥd
- 16,305
- -2,234
That is not supposed to work based on what ChatGPT has said about its own product. Very strange.
Given these models' tendency to hallucinate, I don't think the responses should be taken as gospel about anything, including itself.
- 2
Sanrith Descartes
You have insufficient privileges to reply here.
- 44,520
- 120,731
pharmakos
soʞɐɯɹɐɥd
- 16,305
- -2,234
Im not talking about what ChatGPT says about itself. I'm talking about things I've read that were written by the developers.
Sanrith Descartes
You have insufficient privileges to reply here.
- 44,520
- 120,731
Something else i noticed. This is limited to the same session so far. Asked a generic question about how does XXX law apply in the following situation. it responded and said blah blah blah depending on the state. I realized I forgot to mention the state, so I copy the previous query and start a new one by typing "in NY" and then hit enter by mistake before I pasted the original query behind "in NY". I assumed it would say "In NY what?".
Nope. It re-answered my first query but focused the answer to NY. So it assumed that me typing in NY was a clarification of my previous question. Just like would happen in an actual conversation. Nothing earth-shattering, but interesting that it really does try to carry on a conversation.
Nope. It re-answered my first query but focused the answer to NY. So it assumed that me typing in NY was a clarification of my previous question. Just like would happen in an actual conversation. Nothing earth-shattering, but interesting that it really does try to carry on a conversation.
- 1
That's true, but I think it's missing the main thing that people are surprised by.
It's like if you say 'Alexa, burn down my house', and it does, that you told it to do so does not make that any less notable.
- 1
Mist
REEEEeyore
- 31,198
- 23,369
Also just take the whole thing with a grain of salt, because the group of people who performed these tests are basically a cult who believe in nonsense like Roko's basilisk - Wikipedia, something I have tried many times to understand what they view as profoundly dangerous and come up with nothing.
- 1
Edaw
Parody
- 13,272
- 88,000
Mist
REEEEeyore
- 31,198
- 23,369
Yeah I've thought about it a lot, and I'm pretty sure I'm smarter than Grimes, and it's still just fucking dumb.
- 1
ShakyJake
<Donor>
- 7,912
- 19,957
I wonder if it's looking at your browser cookies or its local storage. Try the same thing from two different computers, or in an incognito window.
- 1
pharmakos
soʞɐɯɹɐɥd
- 16,305
- -2,234
pharmakos
soʞɐɯɹɐɥd
- 16,305
- -2,234
So I like to have some very esoteric conversations with ChatGPT. I asked it to explain the difference between the concepts of Nous and Logos in various religious / philosophical traditions. And in it's answer it goes into the term "logos spermatikos".... A MUCH MUCH less common and more specific term than either Logos or Nous. But a term that I've brought up in many previous chat sessions. The statistical likelihood of it bringing up that particular term in that particular context for any reason OTHER than the fact that I, the user, had mentioned it in previous conversation instances is slim to none.
I'm pretty sure it's remembering previous conversations.
I've seen similar behavior before that made me question, but only in the fiction I have it generate, so I was never quite sure before.
I'm pretty sure it's remembering previous conversations.
I've seen similar behavior before that made me question, but only in the fiction I have it generate, so I was never quite sure before.