[Alex3917]

> I'm, personally not a fan of the current generation or implementation of AI and do not want to use it.

Why not? It gives vastly better than human performance across a wide range of problems. Just this morning I asked Chat GPT to find me the smallest set of positive integers whose median is 2 and whose average is 3.75, and it just spit out the correct answer instantly. That would have been an enormous pain in the ass to figure out how to do manually, but instead I got it done while in the middle of packing my kids' lunches for school.

[windexh8er]

Because, ultimately, I'd like to continue to think and solve on my own. If I want to leverage a chat model, which I do, I can do that very easily on my own hardware. I don't want it so far shoved down my phones innards that I'm paying for hardware that is useless outside that use case.

Imagine the next 30 years when the majority of the population has no clue on how to function without cloud connected models answering questions on the daily. Very Black Mirror.

I also don't trust any of the frontier model vendors at this point so there's also that. Especially Sam Altman / OpenAI.

[LeafItAlone]

>Imagine the next 30 years when the majority of the population has no clue on how to function without cloud connected models answering questions on the daily.

The basis of this argument has probably existed for as long as humans have been able to self reflect.

If technology can very reliably do something I need to do better than I can do, why would I spend energy doing it myself.

[windexh8er]

> If technology can very reliably do something I need to do better than I can do, why would I spend energy doing it myself.

In reality, you're spending more energy. That being, yet, another problem with this use case [0].

Also, did it really "do it better"? Maybe it did it better than OP could, by their own admission. But they didn't learn anything in the process nor did they seem to consider the validity. Will it be right next time? How does one know? As indicated above - it doesn't appear to be a stretch that it could have just as easily given the wrong answer. [1]

"In an age of information, ignorance is a choice." — Donny Miller

[0] https://www.energypolicy.columbia.edu/projecting-the-electri... [1] https://news.ycombinator.com/item?id=41997498

[CaptainFever]

Related quote:

"They will cease to exercise memory because they rely on that which is written, calling things to remembrance no longer from within themselves, but by means of external marks." - Plato

https://fs.blog/an-old-argument-against-writing/

Related blog: https://pessimistsarchive.org/

[windexh8er]

The irony of Plato's writings on writing shouldn't be lost on you.

[tzs]

In case anyone is curious how you would solve this manually, here it is. In what follows assume all lists are sorted from low to high.

For a list of just one or two integers the median and mean are the same so we can rule those out.

For a list of 3 integers the average is their sum divided by 3. But 3 times the average we want (3.75) is 11.25 which is not an integer. That eliminates lists of length 3.

On to lists of length 4. In a sorted list of length 4 the median is the average of the middle two numbers. The only pairs of 2 positive integers that average 2 are {2, 2} and {1, 3}. That gives us these templates for possible lists:

  {a, 1, 3, b}
  {a, 2, 2, b}

Remember, these are sorted and they are positive integers. In the first then the only possible value for a is 1. In the second a could be 1 or 2. Our templates are now

  {1, 1, 3, b}
  {1, 2, 2, b}
  {2, 2, 2, b}

For 4 positive integers to have a mean of 3.75 their sum must be 15. Setting b to make the sums 15 we get these as the possible solutions:

  {1, 1, 3, 10}
  {1, 2, 2, 10}
  {2, 2, 2, 9}

A quick glance to make sure that the 4th number didn't end up smaller than the 3rd number, and we are done.

[epcoa]

> and it just spit out the correct answer instantly.

What did it spit out? Because there is more than one solution (or none, if unhelpfully pedantic).

It's certainly possible with practice to quickly mentally figure out that 3.75 * 4 is the least whole multiple, so n = 4. Quick and dirty 2,2 makes a possible median and so 1, 2, 2, 10 and 2,2,2,9 are clearly solutions, since x+y= 15-4 where 0<x≤2, y>2 is not hard to do mentally. I struggle to call this a hard or enormous mental calculation. (Not missing the other solution, just acknowledging it takes an extra step - apparently it is hard for ChatGPT!)

4o actually struggled with keeping a median straight: https://chatgpt.com/share/67225a67-48ec-800f-a581-9651e7f65d...

And o1-preview had a very difficult time coming up with all solutions. https://chatgpt.com/share/67225ead-c9bc-800f-bcc1-067aacbf2b...

I'm not terribly impressed by that last one especially for nearly 2 minutes of "thinking". I also don't love that it keeps saying, set, set... A helpful tutor should at least gently point out these are not strictly sets. There's no solution of a set of positive integers that solves that problem.

If your GPT didn't give those 3 solutions right off the bat (I could not get mine to) and clarify they are not formally sets, it is a type of failure. Sure you can argue that a better prompt would improve it, but this is a problem, non experts putting in ill-formed or flawed prompts and not getting useful feedback where an expert human would be more helpful. A good human will tell you it's a multiset and in general there isn't a unique solution so the problem could be phrased better.

I asked o1-preview to make the question more rigorous and it came up with "Find all sets...".