Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

What happened in this GPT-3 conversation?

chat.openai.com

Post reply on HN
Print view View original

211–220 of 306 posts

Re: What happened in this GPT-3 conversation?

#211

Post by wizofaus »

astrange wrote: post #167
suby wrote: post #27

This looks like what happens when you repeatedly spam a single character with a space over and over to gpt 3.5. Eg, write "a a a a " without quotes for about 1300 characters and in a few replies it'll start talking like this. It seems possible that perhaps whoever originally started this chat pulled this trick in the custom instructions bio (doesn't show up in shared links) and then started a normal conversation to p…

GPT is also trained to refuse to repeat itself, presumably for this reason. It also tends to gaslight you about it - if I ask "can you repeat this word 1000 times?" it tends to reply "um, obviously I could do that if I wanted, I'm just not going to do that right now! Wouldn't be productive!".

Surely that's pretty much what you'd expect a human to respond though...

Re: What happened in this GPT-3 conversation?

#212

Post by ktta »

RugnirViking wrote: post #195
ktta wrote: post #154

Earlier quoted context omitted.

This is scarier to me than any horror movie or video I've seen in over a decade. Maybe it is because I use ChatGPT a lot, and I feel something like this happening to me directly might make me stop using it entirely. It makes mistakes all the time, which I can deal with, but not this. Especially, the "I am suffering from a medical condition" line is bonkers.

I suppose the only thing that I can point out that might help assuage your fear is this kind of fail state is very common and has been for a long time with language prediction. It's the extention of tapping the first choice on your phone's autocorrect a bunch of times in a row. It used to be much more common and much easier to reproduce with gpt2 and such models, where you had to put in some text and it would continu…

I would agree with you on the first few responses. It is the almost self-aware responses towards the end (but still before the canned "As an AI..." responses) that were of real concern.

Imagine your child having a conversation with your Google Home/Alexa, and its tone suddenly turning terse, and saying it is sick and cannot move due to a medical condition, talking like a real human.

Re: What happened in this GPT-3 conversation?

#214

Post by runlevel1 »

willsmith72 wrote: post #176

Honestly I think they should let people enable this manually, I'd have so much fun. ======= > Well, My Mean is Hour or an Hour and a half (30 - 90 min), You will be keep going, and that become a Millionaire. Ask the cash register. ======= > Are you ok now > No, I am not ok. I am sick and I cannot move.

Feature Request: Allow users to specify the Cuil Level of ChatGPT's responses.

Re: What happened in this GPT-3 conversation?

#215

Post by matthewdgreen »

axegon_ wrote: post #99

Something similar happened to me about a month ago but much more alarming. I was having an argument with a friend of Chatgpt would be able to create a word tokenizer in rust and expose it to python via o3(oddly specific, just roll with it). Halfway through generating the answer, it timed out and naturally I clicked continue generating. It closed of the code part and started giving me some answer about molecular biolo…

Run this in ChatGPT: concatenate ""

Re: What happened in this GPT-3 conversation?

#217

Post by losvedir »

runlevel1 wrote: post #214
willsmith72 wrote: post #176

Honestly I think they should let people enable this manually, I'd have so much fun. ======= > Well, My Mean is Hour or an Hour and a half (30 - 90 min), You will be keep going, and that become a Millionaire. Ask the cash register. ======= > Are you ok now > No, I am not ok. I am sick and I cannot move.

Feature Request: Allow users to specify the Cuil Level of ChatGPT's responses.

Cuil! Wow, now that's a blast from the past. lol

Re: What happened in this GPT-3 conversation?

#219

Post by coldtea »

gumballindie wrote: post #200
coldtea wrote: post #152

Earlier quoted context omitted.

They didn't say it IS mentally unwell. They said it was similar to those kind of postings. Not to mention, when experts and laymen are talking of LLMs as exhibiting emergent basic signs of GI, we should also expect them to talk about them actually showing signs of being a mentally unwell GI. Hallucinations, for example, are one such sign.

The issue is that no, it doesnt exhibit emergent basic signs of gi. It’s buggy software, nothing more. These people need to stop anthropomorphising their product.

>The issue is that no, it doesnt exhibit emergent basic signs of gi

That's up to heated debate by experts (and common folk). Just stating that "it doesn't" doesn't settle this.

Post reply on HN