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

Comparing Google and ChatGPT

twitter.com

Post reply on HN
Print view View original

101–110 of 280 posts

Re: Comparing Google and ChatGPT

#101

Post by telotortium »

paulmd wrote: post #43
mikkergp wrote: post #8

Earlier quoted context omitted.

As everyone starts to adopt AI, are we going to get to a point where the AI is eating itself. I could imagine AI failing similarly to incestuous genetic lines creating mutations.

Yep, as AI starts to get trained on AI-generated data the output may well become unstable, you can't build an infinite motion machine (or an infinite gain machine/infinite SNR amplifier) and the system may degrade to essentially white noise. Sort of a cyber-kessler syndrome basically. You really don't want AI-generated content in your AI training material, that's actually probably not generating signal for building f…

What seems most likely is that OpenAI and other LLM trainers are going to proceed to training on transcripts of YouTube videos and podcasts using the Whisper text-to-speech model, which at its largest sizes is really quite state-of-the-art. For now, it seems like most of this content is still organic (or if it's not, the computer-generated speech is relatively easy to distinguish for now).

Re: Comparing Google and ChatGPT

#102

Post by nullc »

drchopchop wrote: post #6

It's great, until people realize GPT-3 will generate answers that are demonstrably wrong. (And to make matters worse, can't show/link the source of the incorrect information!)

So does google. Always great when you a blurb (or a summary of a link) telling you to drink bleach...

Re: Comparing Google and ChatGPT

#103

Post by bushbaba »

impulser_ wrote: post #88
ben_w wrote: post #42

Earlier quoted context omitted.

I just tried Googling "when did the moon explode?" to see if it still gave authoritative answers to bogus questions: > About an hour after sunset on June 18, 1178, the Moon exploded. "when did lincoln shoot booth" > April 14, 1865 Mostly they seem to catch and stop this now, but there was a fun brief period where it was popping up the fact-box for whatever seemed closest to the search terms, so "when did neil armstro…

You are providing only a small part of the result. If you provide the full result it make prefect sense why Google would suggest it. When you search "when did the moon explode?". The full result is actually > About an hour after sunset on June 18, 1178, the Moon exploded. That's what it looked like to five terrified, awestruck monks watching the skies over the abbey at Canterbury, in southeastern England, anyway. Whi…

Doesn’t seem that crazy for a future iteration of GPT to offer sources.

Re: Comparing Google and ChatGPT

#104

Post by satvikpendem »

I asked on Twitter: "Why do you assume Google (who has one of the largest AI teams around, plus DeepMind) won't also integrate this into search too?"

I mean really, do people really think Google isn't also working on stuff like this?

Re: Comparing Google and ChatGPT

#105

Post by Spivak »

rafaelero wrote: post #99
Tepix wrote: post #16

Earlier quoted context omitted.

Exactly, i talked to ChatGPT and it gave me a lot of wrong information in an authorative tone. I consider it dangerous as-is.

Quoted post unavailable.

I mean it's not like it's dangerous on its own, but if you're like "Hey GPT how do I put out a grease fire?" and it replies "Pour water on it" and you believe it then you're in for a bad time.

So I mean I guess you're technically right, it's not dangerous so long as you have 0% confidence in anything it says and consider it entertainment. But what would-be scrappy Google competitor is gonna do that?

The thing that makes it particularly insidious is that it's going to be right a lot, but being right means nothing when there's nothing to go off of to figure out what case you're in. If you actually had no idea when the Berlin Wall fell and it spit out 1987 how would you disprove it? Probably go ask a search engine.

Re: Comparing Google and ChatGPT

#106

Post by jerf »

adamsmith143 wrote: post #75
jerf wrote: post #50

Earlier quoted context omitted.

Yes they do, and I do not deny the power of human's ability to confidently spew nonsense. However, humans do have some known failure cases that help us detect that. For instance, pressing the human on a couple of details will generally show up all but the very best bullshit artists; there is a limit to how fast humans can make crap up. Some of us are decent at the con-game aspects but it isn't too hard to poke throug…

>Computers can confabulate at full speed for gigabytes at a time. This I think is the actual problem. Online forums will likely be filled with AI generated BS in the very near future, if not already. >"The statistically-most likely conclusion of this sentence" isn't even a poor approximation of truth... it's just plain unrelated. That is not what truth is. At least not with any currently even remotely feasible defini…

"By the same logic how can we allow Humans to do those jobs either?"

First, since you can't see tone, let me acknowledge this is a fair question, and this answer is in the spirit of exploration and not "you should have known this" or anything like that.

The answer is a spin on what I said in my first post. Human failures have a shape to them. You cite an example that is certainly common, and you and I know what it means. Or at least, what it probabilistically means. It is unfortunate if someone with lesser understanding calls in and gets that answer, but at least they can learn.

If there were a perfect support system, that would be preferable, but for now, this is as good as it gets.

A computer system will spin a much wider variety of confabulated garbage, and it is much harder to tell the difference between GPT text that is correct, GPT text that is almost correct but contains subtle errors, and GPT text that sounds very convincing but is totally wrong. The problem isn't that humans are always right and computers are always wrong; the problem is that the bar for being able to tell if the answer is correct is quite significantly raised for me as someone calling in for GPT-based technologies.

Re: Comparing Google and ChatGPT

#107

Post by Toutouxc »

ben_w wrote: post #42
drchopchop wrote: post #6

It's great, until people realize GPT-3 will generate answers that are demonstrably wrong. (And to make matters worse, can't show/link the source of the incorrect information!)

I just tried Googling "when did the moon explode?" to see if it still gave authoritative answers to bogus questions: > About an hour after sunset on June 18, 1178, the Moon exploded. "when did lincoln shoot booth" > April 14, 1865 Mostly they seem to catch and stop this now, but there was a fun brief period where it was popping up the fact-box for whatever seemed closest to the search terms, so "when did neil armstro…

> Is Python even or odd?

> The required code is provided below. num = int (input (“Enter any number to test whether it is odd or even: “) if (num % 2) == 0: print (“The number is even”) else: print (“The provided number is odd”) Output: Enter any number to test whether it is odd or even: 887 887 is odd.

The first sentence is not my query, it's a question that Google offers and answers in their useless box.

Re: Comparing Google and ChatGPT

#108

Post by jerf »

visarga wrote: post #59
jerf wrote: post #50

Earlier quoted context omitted.

Yes they do, and I do not deny the power of human's ability to confidently spew nonsense. However, humans do have some known failure cases that help us detect that. For instance, pressing the human on a couple of details will generally show up all but the very best bullshit artists; there is a limit to how fast humans can make crap up. Some of us are decent at the con-game aspects but it isn't too hard to poke throug…

> Personally, I consider any GPT or GPT like technology unsuitable for any application in which truth is important . Full stop. The technology fundamentally, in its foundation, does not have any concept of truth I think you got it all wrong. Not all GPT-3 tasks are "closed-book". If you can fit in the context a piece of information, then GPT-3 will take it into consideration. That means you can do a search, get the d…

I don't think you're right. Even if you add "correct" context, and in many of these cases "I can locate correct context" already means the GPT-tech isn't adding much, GPT still as absolutely no guard rails stopping it from confabulating. It might confabulate something else, but it still confabulating.

Fundamentally, GPT is a technology for building convincing confabulations, and we hope that if we keep pounding on it and making it bigger we can get those confabulations to converge on reality. I do not mean this as an insult, I mean it as a reasonable description of the underlying technology. This is, fundamentally, not a sane way to build most of the systems I see people trying to build with it. AI Dungeon is a good use because the whole point of AI Dungeon is to confabulate at scale. This works with the strengths of GPT-like tech (technically, "transformer-based tech" is probably a closer term but nobody knows what that is).

Re: Comparing Google and ChatGPT

#109

Post by numpad0 »

tfsh wrote: post #11

These are addressing two very different concerns but framed as a singular one. Google is first and foremost a search engine - it searches the web for answers, the key point being the answers need to exist on the web. The other is a machine learning model tasked with deriving answers, and sometimes - if not very often answers will be provided in an authoritative tone whilst being completely and utterly incorrect. Goog…

I know I’m being idiot on this as always but nor I’m sure why this isn’t said more often: Web search, and by extension Google Search, is a hostile document organization and search system.

Its principle is 1) there is a collection of “stolen Soviet documents”, or the web crawl, 2) obscured slice of meaningful data hidden in it that relates mutually by a $CODEWORD, and 3) “hostile” interest in it from a “spy” overhearing it, that the search engine can then work on to compile into a collection to present.

Whatever that answers a question given it is not a search, it’s something different.

Re: Comparing Google and ChatGPT

#110

Post by adrianmonk »

bccdee wrote: post #53
dougmwne wrote: post #31

Earlier quoted context omitted.

Fair point, but Google is also exactly as confidently wrong as GTP. They are both based on Web scrapes of content from humans after all, who are frequently confidently wrong.

No it isn't. When Google gives you incorrect info, it links the source. GPT-3 will gleefully mash together info from several incorrect sources and share none of them.

If Google is giving you a search result, yes. But Google returns other types of answers, and sometimes they are unsourced and wrong.

For example, do this search:

    who wrote the song "when will i be loved"
The results page contains short section before the web page results. This section says:

    When Will I Be Loved
    Song by Linda Ronstadt
The song was actually written[1] by Phil Everly of the Everly Brothers, who recorded it in 1960. Linda Ronstadt released her version in 1974. Both versions rose pretty high on the pop charts, but Ronstadt's went higher.

But, what does "by" mean -- recorded by or written by? Maybe Google isn't giving me a wrong answer but is just answering the wrong question?

Nope, the Google result also includes a row of pink radio buttons for selecting different info about the song, and the page loads with the "Composer" button selected.

So, it's just plain wrong. And there is no link or other hint where the information came from.

---

[1] https://en.wikipedia.org/wiki/When_Will_I_Be_Loved_(song)

Post reply on HN