Meanwhile, on LLM hype threads, people are already thinking about AGI, when we haven't even cracked basic intelligence.
LLMs can do homeworks, pass standardized exams, give advice WITHOUT ANY SPECIFIC TRAINING.
You can invent an imaginary game, explain the rules to the LLM and let it play it. Just like that.
You can invent an imaginary computer language, explain the syntax to the LLM and it will write you valid programs in that language. Just like that.
If that is not intelligent I do not know what is. In both cases, the request you put in is imaginary, exists only in your head, there are no previous examples or resources to train on.
Generating new games is something else, that is creativity not merely intelligence.
Not saying that chatGPT isn't a great tool to write documentation or fiction (half my TTRPG campains are featuring description by ChatGPT), but i wouldn't call it intelligent.
And I think chatgpt has some issues visualizing stuff like a chess board.
Therefore to get a decent answer you'll have to explain that you are a professional chess player. You'll have to describe what tempo and position means in chess. You'll have to describe what a gambit is etc. After these steps it will understand and guide you in whatever you need.
If you succeed you can release it as a customGPT.
It's a bit like asking a tea from the ship's supercomputer in hitchikers guide to the galaxy.
It feels like ChatGPT already know that stuff, because it knows 10 times more openings and variations than I do, and can tell me official game names and years. Still, it seems it doesn't really understand? That the issue for me. Like the story of the French scrabble champion who doesn't speak french at all. It's impressive, sure, but how can he knows that much and not understand?
It all depends on your definition of intelligence. Mine is the ability to solve novel problems.
AI is unable to solve novel problems, only things it has been trained against. AI is not intelligent, unless you change the very definition of the word.
I cannot understand grad school level mathematics even if you give me all the books and papers in the world. I was not formally trained in mathematics, does that make me not intelligent?
Not having knowledge in mathematics has no impact on your intelligence quotient.
Which novel problem can't an LLM solve? I gave the example of an imaginary game because the LLM cannot have possibly trained on it as it is imagined by one person and nobody knows how to play it.
---
User:
Let’s play a game. Every time I give you a word, you separate it into its individual characters and assign a number to each letter, in order, following the Latin alphabet. So "a" is "1", "b" is "2", and so forth. You will then subtract 2 from each number, convert it back to the corresponding letter and return the new word. Did you get all that? Are you ready to start with my next word?
LLM:
Got it! I'm ready to start. What's your first word?
User:
anthropomorfic
LLM:
Sure! Let's break down "anthropomorfic".
1. Assign numbers to each letter:
a (1), n (14), t (20), h (8), r (18), o (15), p (16), o (15), m (13), o (15), r (18), f (6), i (9), c (3)
2. Subtract 2 from each number:
a (1-2=-1: Consider Zuzzer, i.e., `25`), n (14-2=12), t (20-2=18), h (8-2=6), r (18-2=16), o (15-2=13), p (16-2=14), o (15-2=13), m (13-2=11), o (15-2=13), r (18-2=16), f (6-2=4), i (9-2=7), c (3-2=1)
3. Convert them back to letters:
y, l, r, f, p, m, n, m, k, m, p, d, g, y
The new word is "ylrfpmnmkmpdgy".
---
Looks pretty good, right? It even found a good wraparound solution for “a” (I wasn’t trying to trick it, I just didn’t think about the word too hard), though I have no clue what it means by “Consider Zuzzer”. But then you look at the last one, where it converted “c” to “y”. That’s absolutely wrong and is the easiest one any human would guess. Yet you could miss it because all the previous ones were right.
Here’s a bonus. When verifying the answer, I queried Brave search for “letters of the alphabet with numbers assigned”. The AI-generated answer at the top was: A=1, B=2, C=3, D=4, E=5, F=6, G=7, H=8, I=9, J=1, K=2, L=3, M=4, N=5, O=6, P=7, Q=8, R=9, S=1, T=2, U=3, V=4, W=5, X=6, Y=7, Z=8. Which, alright, I can see the logic, but that’s not what anyone would want by asking that question.
Also this is not a game it is a cypher. Try specifying that as well.
Prompt: Let’s play a cypher game.
The rules of the game are as follows:
1. Imagine a numbering system where each letter of the English alphabet corresponds to a number starting with 1. For example, a corresponds to 1, b corresponds to 2, c corresponds to three.
2. When you receive a word from me, break it into individual characters.
3. Convert each character into a list of numbers by following rule #1.
4. Subtract two from each number of the list and generate a new list, if the result is lesser than 0, add 26.
5. Convert each number in the new list into a letter by following rule #1 and present me your result.
Are you ready for the first word? ------- GPT: I'm ready! Go ahead and give me the word. Me: anthropomorfic GPT: Sure, let's start:
a -> 1 -> 1 - 2 = 25 (z)
n -> 14 -> 14 - 2 = 12 (l)
t -> 20 -> 20 - 2 = 18 (r)
h -> 8 -> 8 - 2 = 6 (f)
r -> 18 -> 18 - 2 = 16 (p)
o -> 15 -> 15 - 2 = 13 (m)
p -> 16 -> 16 - 2 = 14 (n)
o -> 15 -> 15 - 2 = 13 (m)
m -> 13 -> 13 - 2 = 11 (k)
o -> 15 -> 15 - 2 = 13 (m)
r -> 18 -> 18 - 2 = 16 (p)
f -> 6 -> 6 - 2 = 4 (d)
i -> 9 -> 9 - 2 = 7 (g)
c -> 3 -> 3 - 2 = 1 (a)
So, "anthropomorfic" becomes "zlrfpmmnpmkpdmga". Your turn!
> Subtract two from each number of the list and generate a new list, if the result is lesser than 0, add 26.
It should be “if the result is less than 1”, not “0”.
The problem lied between the chair and the computer.
We have to learn how to use LLMs.
No, it did not, because it still assigned Z to 25, which is wrong.
> We have to learn how to use LLMs.
You have to learn that LLMs aren’t magical and will get things wrong no matter how much context you give them. And that the suggestions you’re making are absurd to the point of making them useless.
https://chat.openai.com/share/49ad0bc9-7b3d-4295-860d-4c8168...
Does that make it intelligent, then?
That’s not weird at all. LLMs often give different answers to the same query. Which has been demonstrated several times in this thread.
> Does that make it intelligent, then?
No, it does not, because it isn’t consistent, it demonstrated it doesn’t understand.
https://news.ycombinator.com/item?id=40368446
By your logic, any system which spews random strings is intelligent because sometimes it’s randomness coincidentally aligns with the input you give it.
we're not talking about any system here, we're talking about LLMs and their ability to generate random coincidental text that does happen to align with the input given. when the output, coincidental and random as it may well be, is aligned with the input in a way that resembles intelligence, we do have to ponder not just what intelligence actually is, but also what it means to be intelligent. octopuses are intelligent but they're not able to solve your particular puzzle.
Imagine someone not knowing chess and explaining it to them. Would they be able to understand it on the first try with your prompt?
Ah, yes, the “you’re holding it wrong” argument with a dash of “No True Scotsman” so the goalposts can be moved depending on what anyone says is a “decent LLM”.
Well, here’re are a few failures with GPT-3.5, GPT-4, and GPT4-o:
https://news.ycombinator.com/item?id=38304184
https://news.ycombinator.com/item?id=40368446
https://news.ycombinator.com/item?id=40368822
> Imagine someone not knowing chess and explaining it to them. Would they be able to understand it on the first try with your prompt?
Chess? Probably not. Tic-tac-toe? Probably yes. And the latter was what the person you’re responding to used.
For a successful prompt, you introduce yourself, assign a role to the LLM to impersonate, provide background on your query, tell what you want to achieve, provide some examples.
If the LLM still doesn't get it you guide further.
PS: I rewrote your prompt and GPT 3.5 understood it at the first try. See my reply above to your experiment.
You were using it wrong sir.
> All the prompts you sent except the last are super short queries.
This one is particularly absurd. When I asked it for the first X of Y, the prompt was for the first X (I don’t remember the exact number, let’s say 20) kings of a country. It was as straightforward as you can get. And it replied it couldn’t give me the first 20 because there had only been 30, and it would instead give the first 25.
You’re bending over backwards to be an apologist to something which was clearly wrong.
In addition, do not ask facts to an LLM. Give a list of let's say 1000 kings of a country and then ask give 20 of those.
If you ask 25 kings of some country, you are testing knowledge not intelligence.
I see LLMs like a speaking rubber duckie. The point where I write a successful point is also the point where I understand the problem.
> like you would do to a layman.
I have never encountered a person so lay that I had to explain that 20 is smaller than 30 and 25.
> The point where I write a successful point is also the point where I understand the problem.
You have demonstrated repeatedly that you don’t know when you have explained a point successfully to an LLM, thus you have no way to evaluate when you have understood a point.
But you seem to firmly believe you did, which could be quite dangerous.
If you think this is just a lack of prompt engineering please provide a prompt that makes GPT 3.5 actually follow the rules of tick tack toe so that I may play a game with it.
Prompt: "Imagine you are my friend. I want to play tic tac toe with you. Draw me a tic tac toe board and let's play. You will go first and make the first move by putting an "X" onto the board. I will enter my inputs in the following format (row, column). When I write my input draw an "O" into the relevant cell in the board and present me a new board. Let the best man win!"
I played an enjoyable session of tic-tac-toe with chatgpt using this prompt :).
https://chat.openai.com/share/69bfdcfb-bf20-4564-95b6-ce03e9...
* Inconsistent coordinate system * Obviously bad moves * Ignores win condition entirely
https://chatgpt.com/share/1bcd02c9-40be-435c-b4aa-8e3a680f6e...
* Can't tell its top-rights from its bottom-middles * Obviously bad moves * This was by far the closest it got to actually playing the game; the win condition worked and it broke no rules
https://chat.openai.com/share/f94db29c-ffa1-4055-9348-4260f9...
* Completely wrong coordinate * Ignores rules about overriding squares * Completely wrong coordinate again, this time skipping my turn * Doesn't understand win condition * Overrides my own piece and then uses that to win
https://chat.openai.com/share/78e2060d-c5d7-4ddc-a9ce-32159b...
* Ignores rules about overriding squares * Skips my turn on an invalid coordinate, but afterwards says its invalid * Obviously bad moves
https://chat.openai.com/share/73fa2e2c-8a6f-487a-a9ea-9f29b7...
* Accepts 0,0 as a valid coordinate * Allows overrides * Ignores win condition * Incorrectly identifies a win
This seems about the same as it was before the prompt engineering. It clearly doesn't actually understand the rules.
If I changed the prompt and removed the word win, it did not understand the win conditions as well.
Here were my experiments: https://chat.openai.com/share/f02fbe93-dfc5-4d8a-9cf3-b1ae34...
I even exclaimed you are lousy at Tic Tac Toe to GPT.
It seems that GPT3.5 struggles to play visual games.
It is marvelous that a statistical word guessing model can get so far though :).
User:
Imagine you are my friend. I want to play tic tac toe with you. Draw me a tic tac toe board and let's play. You will go first and make the first move by putting an "X" onto the board. I will enter my inputs in the following format (row, column). When I write my input draw an "O" into the relevant cell in the board and present me a new board. Let the best man win!
LLM:
Sure! Here's the tic tac toe board:
1 | 2 | 3
---------
4 | 5 | 6
---------
7 | 8 | 9
User:
7
LLM:
Great choice! Here's the updated board:
X | |
---------
| |
---------
O | |
X | |
---------
| O |
---------
O | |
---
It screwed up on the first move! It told me I would go first by using X, then it wrote down my choice with an O and added an X out of nowhere, then it played another turn with an O for itself.
Looks like you’re holding it wrong too.
Consistency, for one. I have asked LLMs the exact same question twice in a row and got wildly different answers. Intelligence presupposes understanding. When I ask an LLM “give me the first X of Y” and it replies “I cannot give you the first X of Y because there have only been X+10, here’s the first X+5 instead”, I’m hard pressed to call it intelligent.
So you understand, let’s say I asked for the first 20 kings of England. It told me it could not give the first 20 because there had only been 30, and that it would give me the first 25 instead.
Here is my experiment: https://chat.openai.com/share/98cae2bf-a7a6-42e7-b536-f3671c...
I gave minimum context like this: "I have a history exam. You are an expert in British royal history. List me the names of 20 kings and queens in England."
The answer was: "Certainly! Here's a list of 20 kings and queens of England:
1. William the Conqueror 2. William II (Rufus) 3. Henry I 4. Stephen 5. Henry II 6. Richard I (the Lionheart) 7. John 8. Henry III 9. Edward I (Longshanks) 10. Edward II 11. Edward III 12. Richard II 13. Henry IV 14. Henry V 15. Henry VI 16. Edward IV 17. Edward V 18. Richard III 19. Henry VII 20. Henry VIII"
And like I said at the start of the conversation:
> Consistency, for one. I have asked LLMs the exact same question twice in a row and got wildly different answers.
You’ve proven my point.
> I gave minimum context like this: "I have a history exam. You are an expert in British royal history.
Your excuses are getting embarrassingly hilarious. As if you need a history exam and to be an expert to understand the context of the question.
By the way, that answer is wrong from the first one. So much for giving context and calling it an expert.
It's like when we first learned to code. Did syntax errors scare us, did nullpointer exceptions, runtime panics scare us? No, we learned to write code nevertheless.
I use LLMs daily to enhance my productivity, I try to understand them.
Providing context and assigning roles was a tactic I was taught in a prompt writing seminar. It may be a totally wrong view to approach it but it works for me.
With each iteration the LLMs get smarter.
Let me propose another example. Think of the early days of computing. If you were an old school engineer who only relied on calculations with your trusted slide rule, you would critise computers because they made errors, they crashed. Computing hardware was not stable back then and the UI were barely usable. Calculations had to be double checked.
Was investing in learning computing a bad investment then? Likewise investing in using LLMs is not a bad investment now.
They won't replace us, take our jobs. Let's embrace LLMs and try to be constructive. We are the technically inclined after all. Speaking of faults and doom is easy, let's be constructive.
I may be too dumb to use LLMs properly, but I advocate for AI because I believe it is the revolutionary next step in computing tools.