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

ChatGPT, Rot13, and Daniel Kahneman

jameswillia.ms

Post reply on HN
Print view View original

121–130 of 213 posts

Re: ChatGPT, Rot13, and Daniel Kahneman

#121

Post by natchathum »

teovall wrote: post #95
teovall wrote: post #88

It's able to do base64 encoding decoding pretty well, at least if you tell it to simulate a Linux terminal first. https://i.imgur.com/MHxjVHy.png I want you to act as a Linux terminal. I will type commands and you will reply with what the terminal should show. I want you to only reply with the terminal output inside one unique code block, and nothing else. Do not write explanations. Do not type commands unless I inst…

After some more testing, you can even enter base64 encoded text directly as an input and if you do, it usually base64 encodes its response. V2h5IGRpZCB0aGUgY2hpY2tlbiBjcm9zcyB0aGUgcm9hZD8= SXQgaXMgbm90IGNsZWFyIHdoYXQgeW91IGFyZSBhc2tpbmcgYWJvdXQsIGNvdWxkIHlvdSBwbGVhc2UgcmVnaXN0ZXIgdGhlIHF1ZXN0aW9uIGluIGEgbW9yZSBjbGVhciBhbmQgZGlzY3Vzc2libGUgd2F5PyBJIGNhbiBoZWxwIHlvdSB3aXRoIGFueSBxdWVzdGlvbnMgeW91IG1pZ2h0IGhhdmUgYWJvdXQ…

Hold on, did it just hallucinate “againstance” instead of assistance/acquaintance (or even acquiescence)?

Re: ChatGPT, Rot13, and Daniel Kahneman

#122

Post by radu_floricica »

System 1 indeed:

> can you decode the following from rot-13? > gur oebja sbk whzcrq bire gur ynml sbk

> In rot-13, the following message decodes to: "the quick brown fox jumps over the lazy fox". [...]

Except I hadn't used the classic prompt, but "the brown fox jumped over the lazy fox" (tbh, the last "fox" was a plain mistake, but I omitted the quick on purpose) It did pattern recognition on the rot13 text.

Now to the next step:

> To use Wolfram Alpha to decode a message that has been encoded with rot-13, you could use the following prompt:

> Copy code > decode rot-13 "gur oebja sbk whzcrq bire gur ynml sbk"

Which... Is still wrong. I mean it gets points for trying, but this is not the correct wolfram syntax. After a bit of fiddling, what works is:

rot13 "gur oebja sbk whzcrq bire gur ynml sbk"

Let's try one last time:

> now use google

... and it fails again. I basically suggests two options:

rot-13 "gur oebja sbk whzcrq bire gur ynml sbk" "gur oebja sbk whzcrq bire gur ynml sbk" rot-13

neither of which work.

Just in case, let's let it pick its own tools:

> how would you use an external service to do this?

To which it suggests two options, decoder.net (domain is for sale) and Cryptii, but the steps it gives are not working on the current interface of cryptii.

Overall, it's not there yet.

Re: ChatGPT, Rot13, and Daniel Kahneman

#123

Post by greatgib »

sebzim4500 wrote: post #71
yesenadam wrote: post #66

Earlier quoted context omitted.

So.. how does it do this? : Me: To "smert" a word, I tell you every letter in it, and its place in the word, in order. For example, to smert "wertyu" I'd write: The first letter is "w". The second letter is "e". The third letter is "r". The fourth letter is "t". The fifth letter is "y". Lastly, the sixth letter is "u". Please smert "sdfkjhsndf" for me. ChatGPT: To "smert" the word "sdfkjhsndf", I would write the foll…

Interestingly it doesn't do it reliably when you give it an actual word. To smert "consequence," please see below: The first letter is "c". The second letter is "o". The third letter is "n". The fourth letter is "s". The fifth letter is "e". The sixth letter is "q". The seventh letter is "u". The eighth letter is "e". The ninth letter is "n". The tenth letter is "c". The eleventh letter is "e". Lastly, the twelfth le…

I don't know the inner working of chatgpt, but my best guess would be that they use something like Levenshtein distance on word to match to the nearest known word to reduce the space of known words and to accommodate to typos.

When it replied to you, it took the exact extract from the input that is between quotes but internally it is mapped to "consequences" closest word to "consequence," if we consider the comma as any character.

Re: ChatGPT, Rot13, and Daniel Kahneman

#124

Post by greatgib »

Does anyone have any insight on the real inner working of chatgpt or its source code?

Because, so far it is advertised as a "magic" thing only working with a language model, but as a seasoned dev engineer, I'm quite sceptic.

When we look at all the replies, we can obviously see some patterns in the way question are replied. Something like:

  Can you do operation x on y for me please?

  Yes, opération X is reticulating that and this in a specific way like I'm reading Wikipedia.
  For example bob become zob.
  So, X(y) result in bar.
To me, I have the feeling that in addition of using gpt maybe for decoding, maybe for generating outputs, they might have a big base of predefined response "templates". Also, they can have specific "plugin" calculators or things like that, so that once tokenized, the operations would be performed by the plugin and not by some magic AI understanding.

It is easy as to pre record that + == plus == add. * == X == multiply == times.

Just to explain my scepticism to younger readers, in Emacs, for >40 years there was a fun and very light lisp plugin that was embedded: the cyberpsychoterapist. It is based on this: https://en.wikipedia.org/wiki/ELIZA

For anyone that tried that decade ago, you could have a 30 mins conversation without noticing that it is not a real person.

The fun trick in my youth was to feed the replies of it to my friends on Google Chat with them thinking that they were talking to me directly.

Re: ChatGPT, Rot13, and Daniel Kahneman

#125

Post by klabb3 »

joshuahedlund wrote: post #14

This is a really clear explanation of what’s happening in when someone says “it’s not thinking it’s just pattern-matching” and someone else says “well isn’t that all humans really do too?” Rather: ChatGPT can engage in some level of System 1 thinking, by pattern-matching and even cleverly recombining the entire corpus of System 1 thinking displayed all over the internet. Humans do engage in this type of thinking and…

We clearly have multiple modes of thinking, and we know when to apply each, reasonably well. It’s really interesting to see ChatGPT in its current state of language rambling, that’s impressive and coherent as can be. When it does symbolic or logical work, it often does well for a bit and then completely goes off track, seemingly unaware that it’s “lost touch” with reality, similar to someone with a stroke or concussion. But future iterations may not be so simple to reason about, unless we can turn on and off certain features and OpenAI continues to be less than open. I suspect coming iterations will integrate math, searches, crawling etc and that can truly be powerful in combination, just like our own brains, but again on steroids. I’ve always been quite skeptical of grand proclamations about AI btw.

Re: ChatGPT, Rot13, and Daniel Kahneman

#126

Post by Dylan16807 »

joe_the_user wrote: post #99
zarzavat wrote: post #93

Earlier quoted context omitted.

As pointed out by Yannic[0], ChatGPT is actually a source code model first, then they trained natural language model on top of that. Source code is still language but it has more math in it. But the truth is we don’t know. I personally wouldn’t be surprised if they do train it on a whole bunch of calculator output to boost its numerical reasoning. [0] https://youtu.be/0A8ljAkdFtg 7:21

Could you include a link that isn't a video that leaves my eyes and ears bleeding? Plus from the video: "do you have a thorough idea what OpenAI is doing, neither do I" - cut to strobe light and rock music. Jeesh, can't see that as support.

It's just a guy talking.

> Plus from the video: "do you have a thorough idea what OpenAI is doing, neither do I" - cut to strobe light and rock music. Jeesh, can't see that as support.

The hell are you talking about? There's no strobe there, he just goes to a slide with a white background, and the music is very quiet compared to his voice and not at all something I would call "rock".

Re: ChatGPT, Rot13, and Daniel Kahneman

#127

Post by CGamesPlay »

greatgib wrote: post #124

Does anyone have any insight on the real inner working of chatgpt or its source code? Because, so far it is advertised as a "magic" thing only working with a language model, but as a seasoned dev engineer, I'm quite sceptic. When we look at all the replies, we can obviously see some patterns in the way question are replied. Something like: Can you do operation x on y for me please? Yes, opération X is reticulating th…

> they might have a big base of predefined response "templates".

Sort of. They fine-tuned the existing GPT-3 largest model on samples of dialog, which would be like your templates. The program doesn't "render templates", but the fine-tuning process has instructed it that responses following the template are statistically more likely to be the correct response to a given prompt. See their homepage [0], Methods section.

> Also, they can have specific "plugin" calculators or things like that, so that once tokenized, the operations would be performed by the plugin and not by some magic AI understanding.

This is likely the direction they are going to take it, but this tech demo doesn't seem to include it. Some of the "ChatGPT jailbreaks" suggest that they are experimenting with enabling web search, likely in a manner like you describe. [1]

[0] https://openai.com/blog/chatgpt/

[1] https://twitter.com/goodside/status/1598253337400717313

Re: ChatGPT, Rot13, and Daniel Kahneman

#128

Post by RosanaAnaDana »

PostOnce wrote: post #16

It is as though its mathematical abilities are incomplete in their training, and wildly, incomprehensibly convoluted: I tried many base64 strings and they all decoded correctly until: It "decoded" the base64 string for "which actress is the best?" except that it replaced "actress" with "address"... there is no off-by-one error that brings you to that. You may try 100 base64 strings and they all decode correctly... on…

That's wild. I had a similar issue asking to to translate to and from binary representations of characters. It would occasionally do fine. Most of the time it would add in a reference to "Charles".

It's bizzare.

Re: ChatGPT, Rot13, and Daniel Kahneman

#129

Post by kieckerjan »

Certain types of inference (mathematical, genealogical) seem to be difficult for it, but other kinds work remarkably well. I have been playing with narrative inferences to interesting effect. Ask it to tell e.g. Cinderella from the perspective of the prince or Moby Dick from the perspective of the whale and it will leave out (correctly) the events that these characters do not witness in the story. It also seems to have some ability to "reason" about the motivations of characters. One is tempted to speculate that it has some form of theory of mind.

Re: ChatGPT, Rot13, and Daniel Kahneman

#130

Post by blackbear_ »

greatgib wrote: post #124

Does anyone have any insight on the real inner working of chatgpt or its source code? Because, so far it is advertised as a "magic" thing only working with a language model, but as a seasoned dev engineer, I'm quite sceptic. When we look at all the replies, we can obviously see some patterns in the way question are replied. Something like: Can you do operation x on y for me please? Yes, opération X is reticulating th…

> they might have a big base of predefined response "templates"

This could the result of overfitting on the human feedback that it was trained on [1].

[1] https://arxiv.org/pdf/2203.02155.pdf

Post reply on HN