ChatGPT knows things that Google doesn't | Hacker News Reader

ChatGPT knows things that Google doesn't

alexmolas.com

12 points_xivi17 comments

Seems like (at least to stack exchange) it's not true:

https://literature.stackexchange.com/questions/26108/did-vol...

(Looks like it was asked by the author)

I just tried the exact same query on GPT4. Here are the results. [1]

> The quote, "I'm not a believer, but I prefer my barber to be a Christian, especially when he's using his razor on my neck," is attributed to Napoleon Bonaparte. This statement reflects Napoleon's pragmatic approach to religion, indicating his preference for a Christian barber, perhaps due to a belief in the moral framework provided by Christianity, which he might have seen as a safeguard while in a vulnerable position. It's important to note that the exact wording of historical quotes can vary due to translations and the passage of time.

https://imgur.com/a/nq2LHr0

when you have odd or obscure queries like this I find it good practice to only provide half the quote, ask GPT to finish it then provide the author. In my case I got the reply (note, it says it cannot find, but it has no internet access)

> I'm sorry, but I couldn't find a quote that matches exactly with what you provided. It's possible that the quote might be misremembered or misattributed. Could you provide more context or clarify further?

where after some chatting, hypothesizing, ruminating and general conversation we got to

> ah, I see what you're referring to now. The quote you're thinking of is attributed to Voltaire, a French Enlightenment writer, historian, and philosopher. The quote goes: "I want my lawyer, my tailor, my servants, even my wife to believe in God, because it means that I shall be cheated and robbed and cuckolded less often. … If God did not exist, it would be necessary to invent him." This quote encapsulates Voltaire's complex views on religion, where he saw the societal utility of belief in God, even if he himself was skeptical.

if you google that quote, you get many hits. While napoleon has said similar things about religion, he isn't attributed with this specific quote either. I think the quote the original article is searching for was a paraphrasing of the two, or perhaps a third yet not discussed author. Point is, if you provide all the information you have you can't verify anything, its better practice to provide incomplete information and then compare against what you do know to try get a feel for how much bullshitting is going on.

I found ChatGPT to be particularly good at these kinds of things - in most cases, it just knows what I'm talking about, despite very vague hints.

And that part just blows my mind. Sure, LLMs can do fancy things these days, but the fact that a model, stored in some disk in a finite space, disconnected from the index of the web, can know so much about so many things blows my mind.

It doesn’t know.

“Knows” is such a useless word now; we have debased it by using it loosely in non-AI code contexts so now we can’t sensibly debate whether a non-sentient thing can “know” anything.

This here is a wonderful perspective. We are most definitely lacking a serious set of epistemology discussions for some time even before the bullshit faucets.

On seeing the link test I automatically amended it to

> ChatGPT "knows" things that Google doesn't

reading through it seems like that was probably correct. That said OpenAI could have fed it a book that's not on the internet. However I'm not sure why this guy didn't ask the bot for a source! Not that I'm bothering to do that either ;)

It might have just made it up, though. It’s very good at making up source references.

I mean, you can treat ChatGPT's answer as a new hypothesis for which you now might have additional tools available to verify.

I.e. before, you had to Google "who is this quote from" which didn't give any results, but now you can do more narrowed queries to test the hypothesis that it's from Voltaire - e.g. seach for Voltaire quote collections and see if its in there, etc.

they both make the same dumb mistake, but not analysing or error checking their own answers.

Great example here that I did:

Whats the next number in this sequence: 71 64 58 53 48 44 ...

Bard gets it right, but for the wrong reason, then gets it even further wrong

The next number in the sequence is most likely 40. This is because the sequence appears to be decreasing by 7 each time.

Here's the breakdown:

71 - 7 = 64 64 - 7 = 57 57 - 7 = 50 50 - 7 = 43 43 - 7 = 36 36 - 7 = 40

Therefore, the next number in the sequence would be 40.

However, it's important to note that without further context or information about the origin of the sequence, there could be other possibilities. For example, the sequence could be part of a more complex pattern or formula.

If you have any additional information about the sequence, I might be able to provide a more accurate prediction for the next number.

ChatGPT 3.5 gets it right for the wrong reason but its working out is for the right reason. To determine the next number in the sequence, let's analyse the differences between consecutive terms:

71 - 64 = 7 64 - 58 = 6 58 - 53 = 5 53 - 48 = 5 48 - 44 = 4

It looks like the differences are decreasing by 1 each time. Following this pattern:

44 - 4 = 40

So, the next number in the sequence is 40.

To show the problem here is an extrapolation. I asked it to tell me what the next 3 numbers would be The given sequence is decreasing, and the difference between consecutive terms is also decreasing. Let’s look at the differences:

71 - 64 = 7 64 - 58 = 6 58 - 53 = 5 53 - 48 = 5 48 - 44 = 4

The differences are not consistent, but we can see that they decrease by 1, then stay the same, and then decrease by 1 again. If this pattern continues, the next differences should be the same as the last one, which is 4, then decrease by 1 to 3, and then stay the same at 3.

So, the next three numbers in the sequence would be:

44 - 4 = 40 40 - 3 = 37 37 - 3 = 34

This is wrong again, the next three numbers would be: 44-4 = 40 40-4 = 36 36-3 = 33

This is why Google is terrified of chatgpt and was caught with the proverbial pants down.

Microsoft should bring back ask Jeeves as an AI chat bot.

You missed the most likely hypothesis: you hallucinated the quote, and ChatGPT hallucinated the attribution.

ChatGPT didn't hallucinate the quote, because you provided it.

I think it’s from a movie but I don’t remember which one.

It's probably Mandela Effect from pondering this too long, but I could swear it was said by Al Swearengen on HBO's Deadwood, which would be pretty amusing if true (although likely not).

The problem with this is that semantically the quote the author is looking for matches this

> I want my lawyer, my tailor, my servants, even my wife to believe in God, because it means that I shall be cheated and robbed and cuckolded less often. … If God did not exist, it would be necessary to invent him

which is far closer to the words typically attributed to Voltaire. So the title "chatGPT understands queries google doesn't" might be mroe correct because if you google the quote i've provided it is attributed correctly without any real issues.

Reply on news.ycombinator.com