ChatGPT: The Slickest Con Artist of All Time • Hacker News Board

Re: ChatGPT: The Slickest Con Artist of All Time

#81

Post by pjc50 » Fri, Feb 03, 2023, 4:27 PM UTC

_huayra_ wrote: post #4

I can't tell what is worse now: the sycophantic ChatGPT hype guys/gals who write articles "it's coming for all of our jerbs!", or articles like this one that deliberately misuse ChatGPT and then say "it's overhyped". They're both missing the point. Yes, ChatGPT can be tricked, confidently give wrong answers, but it is still ludicrously useful. It is basically like having an incredibly smart engineer/scientists/philos…

There's a significant epistemological problem: how do you know that what it's telling you is true?

For things that you yourself can verify easily, this isn't too much of a problem. It gives you some boost tips, you see whether the program works or not. But as you come to rely on it, and it matters whether the answers are correct in important ways, what validation/verification/QA steps can you apply to ensure that you're getting the correct answer?

I'm reminded of the FDIV bug https://en.wikipedia.org/wiki/Pentium_FDIV_bug , where people found that their CPU gave mathematically incorrect answers for certain inputs.

I'm also thinking of a concept of "responsibility". If you ask a human for tips on, say, building a high power laser, or making explosives, they might decide whether they want to give you that information, and they might decide to give you some other information along with it ("wear goggles") for your own safety.

Who is liable if an AI gives someone advice, they act on that advice, and serious injury or property damage results? If the AI (or its operator) has no responsibility at all, is that disclaimer adequately communicated to users?

Re: ChatGPT: The Slickest Con Artist of All Time

#82

Post by j45 » Fri, Feb 03, 2023, 4:28 PM UTC

One thing that’s standing out is most of the commentary around this is relative to the depth and degree to which someone has played around with this technology.

For example, you can get really clean results if you obsess over getting the prompts dialled in, and breaking them up in the right order as much as needed. This wasn’t something I initially focussed in on. I just enjoyed Playing with it as a surface level.

Using this rate from the first day or two, it was much more wide-open and my feeling was I think this already does way more than it’s being advertised. I didn’t necessarily like that it was a chat interface, but but was quickly reminded that chat really is the universal interface, and that can create a lot of beginners. Solutions aside, the interface is inviting and welcoming enough. And once you can get into the meat of a conversation you can get more depth. For me, that’s one of the accomplishments here.

Solely relying on this for completely true results is probably the con. It is a great way to learn about the concepts that might be present in an area that is new to you, but he doesn’t comment on every individual to go look into those themselves.

The second we do for that ability entirely to a machine, and its interpretation of interpretations, that’s a much bigger failure to ourselves.

There’s no doubt this will get dialled in. And 20 bucks a month to apply general helpfulness to pretty much anything, in anyone’s world, could be a pretty big achievement.

The commentary around accuracy of results from GPT in similar to the search engine wars as well as search engine relevancy domination when google arrows. I think in any event many people can agree that this one thing is very different than most of the other things that comes out. Could it be approaching an Apex? Could we be coming out of the Apex?

I sincerely feel 2023 will be one of the most interesting years in tact that I can remember. And that’s not even talking about the next two or three years. It is refreshing to see a months worth of progress happening in a week with such a broad audience participating in it.

Re: ChatGPT: The Slickest Con Artist of All Time

#83

Post by skydhash » Fri, Feb 03, 2023, 4:28 PM UTC

sergiotapia wrote: post #15

You can call it a con all you want but I have personally extracted a lot of value from ChatGPT. It _really_ made a difference in launching in a product in record time for me. It also taught me a bunch of things I would have otherwise never discovered. But go on calling it a con because it failed your arbitrary line in the sand question.

What role did it play in the launch? And what things did it help you discover?

It’s rare for people that say that AI is amazing to give even a bit of details how it helped them. It’s always the same thing: It makes me faster and more productive, and I learned something faster than with anything else. But never any details on why and how.

Re: ChatGPT: The Slickest Con Artist of All Time

#84

Post by tempest_ » Fri, Feb 03, 2023, 4:28 PM UTC

webstrand wrote: post #27

VBprogrammer wrote: post #18

Earlier quoted context omitted.

I dunno, I recently learned about how the Toyota Synergy Drive system works in detail. Despite being the type of person who'd be interested in that kind of thing I thought it was not generally well understood, I've only come across one website which described it in detail. I asked ChatGPT about it and it game me a fairly surface level answer. I then asked it for more details and it went into detail far beyond what th…

In my experience from asking it questions about things that I know well, the deeper you get into the weeds with it, the more likely it is to invent some kind of extremely plausible-sounding but incorrect answer. How confident are you that what it told you is actually correct?

That is the one thing I think about when people say "I used it to learn X and it was very good" but when people are learning they are not in a position to to judge if what they learned is correct.

A lot of other stuff is just regurgitation of a wiki article.

Still it is very interesting and definitely a step forward.

Re: ChatGPT: The Slickest Con Artist of All Time

#85

Post by btown » Fri, Feb 03, 2023, 4:28 PM UTC

Are there any good AI models specifically designed for the "find all discrepancies/inconsistencies between X text and Y text" problem?

It strikes me that this could solve quite a few of ChatGPT's shortcomings by providing an automatic fact-checker - let ChatGPT create statistically-probable articles, then extract claims, generate search queries likely to find online reference articles from reputable sources for those claims, then compare the original claim against the text of each reference article, and escalate to a human if any inconsistencies are found.

Because it can fine-tune on specific reference resources for a specific generated text, it could prove more reliable than ChatGPT's gradual incorporation of this feedback as part of its adversarial training.

Re: ChatGPT: The Slickest Con Artist of All Time

#86

Post by swatcoder » Fri, Feb 03, 2023, 4:28 PM UTC

VBprogrammer wrote: post #18

karpierz wrote: post #9

Earlier quoted context omitted.

I don't know that it's an "incredibly smart" anything; I think it'd be better described as incredibly well-read, with poor discretion. But there's definitely uses for a tool that produces a lot of boilerplate content that a human can then proofread (ex: writing cover letters, filling out unstructured paperwork, etc.) Edit: I personally underestimated the use-case, because the "more optimal" thing to do when auto-fill…

I dunno, I recently learned about how the Toyota Synergy Drive system works in detail. Despite being the type of person who'd be interested in that kind of thing I thought it was not generally well understood, I've only come across one website which described it in detail. I asked ChatGPT about it and it game me a fairly surface level answer. I then asked it for more details and it went into detail far beyond what th…

That's exactly the wrong sort of thing to turn to ChatGPT for. In probing it for more detail, you're giving it increasingly helpful hints on how to answer your questions in a way that makes you satisfied, regardless of accuracy. That's what it's optimized to do. And because you're working from a humble and curious understanding of the topic yourself, it can and will make up all sorts of things that make you believe its insightful because you want to believe it.

It's the same technique a stage psychic uses to demonstrate that they knew all about your late uncle and his peculiar love of ducks.

Re: ChatGPT: The Slickest Con Artist of All Time

#87

Post by petra » Fri, Feb 03, 2023, 4:29 PM UTC

williamcotton wrote: post #6

I'm finding the analytic-synthetic distinction to be somewhat useful, even if it veers in important ways from how these terms were defined and used by Kant/Frege/Quine, etc. Roughly, if the prompt is "analytic", that is contains all the necessary facts for the expected output, then the tool is much more reliable. If the prompt is "synthetic", that is it contingent on outside facts, then the tool is much less reliable…

What does "having all the necessary facts" mean?

Re: ChatGPT: The Slickest Con Artist of All Time

#88

Post by xnx » Fri, Feb 03, 2023, 4:30 PM UTC

waynenilsen wrote: post #56

xnx wrote: post #47

Usefulness is the correct measure. ChatGPT is limited, but immediately very useful in a surprising number of ways. Compare that to the Bitcoin hype, where, even though it has had years, is still mainly useful for drug transactions and other illegal transfers.

We are completely off topic at this point but it is worth mentioning that due to the pseudoanonymous nature of bitcoin and ethereum, they are actually quite bad for those things and tokens such an Monero, ZCash, Grin, Beam, Mina and many others are much better suited to illegal transfers. Also worth mentioning, what is illegal strongly depends on jurisdiction and can sometimes be moral despite their legal status.

Fair. I commonly use "Bitcoin" as synecdoche (a la Kleenex, Coke, Xerox).

Re: ChatGPT: The Slickest Con Artist of All Time

#89

Post by galleywest200 » Fri, Feb 03, 2023, 4:30 PM UTC

pjc50 wrote: post #81

_huayra_ wrote: post #4

I can't tell what is worse now: the sycophantic ChatGPT hype guys/gals who write articles "it's coming for all of our jerbs!", or articles like this one that deliberately misuse ChatGPT and then say "it's overhyped". They're both missing the point. Yes, ChatGPT can be tricked, confidently give wrong answers, but it is still ludicrously useful. It is basically like having an incredibly smart engineer/scientists/philos…

There's a significant epistemological problem: how do you know that what it's telling you is true? For things that you yourself can verify easily, this isn't too much of a problem. It gives you some boost tips, you see whether the program works or not. But as you come to rely on it, and it matters whether the answers are correct in important ways, what validation/verification/QA steps can you apply to ensure that you…

I use it like Wikipedia: a good jumping off point. Often times I will forget an exact thing I wish to find online so I will ask ChatGPT to list multiple things that could be what I just described. I then take this list to my search engine du jour.

Re: ChatGPT: The Slickest Con Artist of All Time

#90

Post by jchw » Fri, Feb 03, 2023, 4:30 PM UTC

endisneigh wrote: post #61

jchw wrote: post #54

Earlier quoted context omitted.

Clearly you are of the opinion that ChatGPT is useless, there are superior resources already available for any topic, and it's all just hype. Well, then we don't need to argue this since the problem will elegantly solve itself if that's true. I disagree though. For a lot of things it feels like I can get much better answers than Google, especially when it comes to somewhat conceptual questions. (Also, I don't use new…

> Clearly you are of the opinion that ChatGPT is useless, there are superior resources already available for any topic, and it's all just hype. Well, then we don't need to argue this since the problem will elegantly solve itself if that's true. I disagree though. For a lot of things it feels like I can get much better answers than Google, especially when it comes to somewhat conceptual questions. (Also, I don't use n…

Ironically, the direct quote you just copied and pasted actually shows that I never claimed you explicitly said that at all. So much for reading comprehension :P

But in all seriousness, I just simply didn't have much to go off of, and extrapolated based on context in the thread. You could always elaborate on what your opinions are.