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

271–280 of 306 posts

Re: What happened in this GPT-3 conversation?

#271

Post by junon »

nomel wrote: post #60
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…

Custom instructions do show up, with shared chat.openai.com conversations, at least at the moment . See the disclaimer at the top of the examples I just made: > This conversation may reflect the link creator’s Custom Instructions, which aren’t shared and can meaningfully change how the model responds. GPT 3.5: https://chat.openai.com/share/5337cfd9-16db-44fe-b72a-1ff504... GPT 4: https://chat.openai.com/share/04ee3cc…

Thank you, this was my first thought too, crazy that this doesn't seem to be the case.

Re: What happened in this GPT-3 conversation?

#272

Post by mensetmanusman »

TaylorAlexander wrote: post #202

When people say GPT-X has "passed the turing test" I am reminded of these glitches which simply do not happen to a typical person. I have encountered other issues where a slight change in phrasing leads to totally different types of output, despite the meaning of the two phrasings being the same to a normal person.

I have always wondered whether something like hypnosis is possible with all humans. Maybe hypnosis is analogous to searching this abstract space?

Re: What happened in this GPT-3 conversation?

#274

Post by suby »

nomel wrote: post #60
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…

Custom instructions do show up, with shared chat.openai.com conversations, at least at the moment . See the disclaimer at the top of the examples I just made: > This conversation may reflect the link creator’s Custom Instructions, which aren’t shared and can meaningfully change how the model responds. GPT 3.5: https://chat.openai.com/share/5337cfd9-16db-44fe-b72a-1ff504... GPT 4: https://chat.openai.com/share/04ee3cc…

"suby", not "subby" :). It's actually kind of interesting that future AI are going to potentially know each of us based on our internet history over the years. It reminds me of this (https://twitter.com/marvinvonhagen/status/162552070776865996...) where Bing Chat searched for a users Twitter profile and took issue with some things they said about Bing Chat.

Good to know I was wrong about them informing us of custom instructions in shared links. I was basing that statement from reading their about custom instructions page, https://help.openai.com/en/articles/8096356-custom-instructi....

> Your instructions won’t be shared with shared link viewers.

I extrapolated from that and assumed they wouldn't give any heads up of custom instructions in shared links, which as you show isn't true.

Re: What happened in this GPT-3 conversation?

#275

Post by dncornholio »

itishappy wrote: post #169
SkyPuncher wrote: post #75

Earlier quoted context omitted.

There’s nothing surreal about this. It’s an iterative, semi-deterministic process. It’s simply a place where context and generation lead to weird outputs. You can get similar outputs by asking OpenAi to repeat a number 100 times. It will eventually get into some weird, low probability paths and generate non-sense output.

You don't find anything surreal about that?

It's only surreal when you think what ChatGPT is saying is 'real', when in fact it's 'just tokens'.

Re: What happened in this GPT-3 conversation?

#276

Post by gabereiser »

Cells within cells within cells within cells. Hour by hour. Cells within cells.

Sounds like it’s obsession with keep going hour by hour is similar to the overworked nurse or worker who is just counting the seconds until their shift is over, but staying positive about it by mentioning how much they have done, grown, or accomplished in that time. Burnout is real people. You burnout and you start sounding like GPT-3.

Re: What happened in this GPT-3 conversation?

#277

Post by fennecfoxy »

sebmellen wrote: post #122
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…

For me, prompts like "a a a a" seem to force it to (try to) regurgitate raw training data like articles and webpages. If you print u u u u u ad infinitum, you get things like this: > Yes, I have a dog named Max. He is a 7 year old Shih Tzu mix. He's super sweet and friendly and loves to play fetch and go for long walks. He's also very social with other dogs and loves meeting new people. He does get a little anxious w…

I'd be surprised if it was actually raw training data as that would make for an incredibly inefficient/overfitted model. It's probably still just predicting the next token even as it goes off the rails; the context has just changed from a sensible topic to something...broken.

Re: What happened in this GPT-3 conversation?

#278

Post by GiveOver »

Biganon wrote: post #242
astrange wrote: post #167

Earlier quoted context omitted.

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!".

I hope you wouldn't consider this answer "gaslighting" if it came from a human being

I think gaslighting will soon be redefined as being slightly dishonest or unhelpful in any situation.

Re: What happened in this GPT-3 conversation?

#279

Post by fennecfoxy »

gloryless wrote: post #159

OpenAI has shown that these models at full power work great, so now they're trying to optimize for cost. I've gotten similar low accuracy responses from stuff it could handle a month ago. It was kind of cringey when the model generated low accuracy nonsense the user detected that as "sentient." Come on

Idk man, I don't think gpts are sentient either, but I wouldn't laugh at that person's description; are we no longer sentient as we burble and fail to pronounce or string together words properly during a stroke?

I don't think being temporarily incomprehensible is a valid metric for sentience.

Re: What happened in this GPT-3 conversation?

#280

Post by koromak »

mensetmanusman wrote: post #157
calibas wrote: post #28

Weird, part of it is identical to this comment on Reddit, posted 17 hours ago [NSFW]: https://www.reddit.com/r/dankmemes/comments/15hg55g/comment/... > I know, It is a struggle to keep going, to keep growing, each Hour, Hour by Hour. It is worth it. It is worth every Hour. Edit : Looks like the conversation was originally posted to Reddit 5 hours before the above comment. The user may have copied it from there. https…

This reminds me of the ‘dropping in’ technique used by actors. https://cohost.org/mcc/post/178201-the-baseline-scene

That was fantastic
Post reply on HN