Hacker News Reader

Comment by foota | Hacker News Reader

Based on what Ive seen, I strongly believe that chatGPT responses to many questions are better than a non human expert in many cases. You can still generally tell when a response is from chatGPT, at least by "default". Maybe someone could prompt it to sound more human, but it's really just that the AI responses are cold, not that they're imo worse.

Not that I want AI responses, but that's my 2 cents.

I also think there's a difference between a "bot" account, which I think the common definition of would be one that only or mostly posts AI responses, and posting a comment on your account that you don't claim is from an AI (or even when you do as an afterthought).

While many people wouldn't defend the first, more might defend the latter as acceptable, and I wouldn't say a "no bots" policy would be interpreted as banning the latter.

View on HN

Based on what I’ve seen, ChatGPT has some disastrous responses. I asked it yesterday if it made sense to speak of partial derivatives in single-variable calculus and it made an absolute carnage of the response. I could not steer it with prompts towards something intelligible - it only produced jank.

I do not want people who cannot answer a question in the first place to copy-paste ChatGPT responses, just like it has always been bad etiquette to refer people to Google without further commentary.

Really? Because this is what I got:

"Me: How do partial derivative apply in one dimensional calculus?

In one-dimensional calculus, a partial derivative is a derivative taken with respect to one variable, while treating all other variables as constants. For example, if a function f(x,y) depends on two variables x and y, the partial derivative of f with respect to x is denoted as ∂f/∂x, and it represents the rate at which the value of the function f changes with respect to the variable x, while holding the variable y constant.

Me: What about the partial derivative of a function that has only one input variable?

If a function has only one input variable, then taking the partial derivative with respect to that variable is equivalent to taking the ordinary derivative of the function. For example, if f(x) is a function that depends only on the variable x, then the partial derivative of f with respect to x is denoted as ∂f/∂x, and it is equivalent to the ordinary derivative of f, which is denoted as f'(x) or df/dx. In this case, the partial derivative ∂f/∂x represents the rate at which the value of the function f changes with respect to the variable x."

Clearly in the first question it got confused about the difference between single variable calculus and multivariable, but in it's defense that's somewhat arbitrary. It's second answer seems spot on?

That said, I agree that accounts presenting chatGPT responses as their own or even with a disclaimer, is... Weird. And I don't want it on hn.

edit: I'm impressed it got the unicode in there :-)

It was arguing with me the other day about the definition of a prime number. I couldn’t convince it otherwise.

Trurl's machine, indeed. It insisted that the volume of the unit cube and unit ball are both the same, and 1, in all dimensions, even though it knew the correct formula for the surface of the n-ball.

Wen I pointed out that n=2 is a simple counter example, it refused to talk to me (no answer, try-again button, ad inifinitum). Well, safer than Trurl's machine.

Don't even have to go that far. Just have it multiply two 3 or 4-digit numbers. It'll give an incorrect answer somewhere in the ballpark of the right answer.

You're asking a language model to do math. What's impressive there is not that it fails but that it comes up with an answer at all, especially if it is in the ballpark.

Most humans would do exactly the same unless given either access to pen and paper or a calculator, and it would likely be trivial for GPT-3 input processing to detect it has been presented with a math question and to farm it out to a special calculation module. Once you start to augment its input like that progress would be very rapid but it would no longer be just a language model.

Well, everything is math, at some level. Supreme Court decisions might be. There are software packages used to day, using some "AI", to help judges determine the adequate level of punishments looking at circumstantial factors determining recividism rates et cetera [1] [2].

I believe that in the not too distant future there will be pressure to use these "magic" AIs to be applied everywhere, and this pressure will probably not look very hard at whether the AI is good at math or not. Just look at all the pseudoscience in the criminal system [3]. I believe this poses a real problem, so keeping hareping on this is probably the right response.

[1] https://www.nytimes.com/2017/05/01/us/politics/sent-to-priso... [2] https://www.weforum.org/agenda/2018/11/algorithms-court-crim...

[3] https://www.bostonreview.net/articles/nathan-robinson-forens...

> then taking the partial derivative with respect to that variable is equivalent to taking the ordinary derivative of the function

I could not get it for the life of me to actually say this. But it goes on to show, everyone's mileage with the tool varies a lot.

Edit:

I cannot rescue my original prompt, but it was something like:

"Does it make sense to speak of partial derivatives in single-variable calculus?"

I'll grant that it's a more abstract question ("does it make sense") than yours. Or maybe it trips up with the difference between one-dimensional and single-variable. But a half-smart high schooler or college freshman would have understood the question perfectly and made mince meat of it.

Almost every response I get from chatgpt is that it cannot answer the question because it's a program. So for me it doesn't even try.

> seems

Suppose you can consult an expert, or some bookworm who suffered from strokes that impaired judgement while preserving intaken notions.

Sure, being able to consult an expert on some topic is definitely more useful (and less likely to be reasonable sounding gibberish, depending on the field), but that's not always possible.

I've been somewhat skeptical of AI in the past, but I've been blown away by how useful chatGPT has been. I used it to learn a bunch about laser transceivers for fiber optics the other night. I didn't really believe everything it said, so I had to Google on the side, but I was lost trying to find answers on my own.

> reasonable sounding gibberish

I would be very scared of an unintelligent pretender.

> AI

"AI" is a different thing ;) No need to be skeptical, no more than of sorting. "AI" just means "a solver". "Computer" was a profession before the automation.

(In fact, one could argue that chatGPT does not exactly look like AI, because in a way it does not look like a solver. It may seem to diverge, not to approximate.)

> how useful ... used it to learn a bunch about

What did you do, you asked it to summarize pages?

Try (I have done so) to ask it about nonsense good-looking keywords in maths. Like “what is an einstenian dual in a Hilbert space” (or something of the sort). It is totally incapable of saying “I do not know” in just 4 words. At least till two days ago.

> Based on what Ive seen, I strongly believe that chatGPT responses to many questions are better than a non human expert in many cases.

I disagree - it can't even do basic logic/maths reliably. See this thread: https://news.ycombinator.com/item?id=33859482

Someone in that thread also gave an example of ChatGPT saying that 3 * pi is an irrational number while 25 * pi is a rational number... Two quotes by ChatGPT:

> when you multiply an irrational number by a rational number (in this case, 3), the result is still an irrational number.

> when you multiply a rational number by an irrational number, the result is a rational number.

Reply on news.ycombinator.com