Sean Caroll has talked about a few word puzzles he asked chatgpt and gpt4 or whatever and they were interesting examples. In one he asked something to the effect of “if i cooked a pizza in a pan yesterday at 200 C, is it safe to pick up?” and it answered with a very wordy “no, its not safe” because that was the best match of a next phrase given his question, and not because it can actually consider the situation.
I don’t deny that this kind of thing is useful for understanding the capabilities and limitations of LLMs but I don’t agree that “the best match of a next phrase given his question, and not because it can actually consider the situation.” is an accurate description of an LLM’s capabilities.
While they are dumb and unworldly they can consider the situation: they evaluate a learned model of concepts in the world to decide if the first word of the correct answer is more likely to be yes or no. They can solve unseen problems that require this kind of cognition.
But they are only book-learned and so they are kind of stupid about common sense things like frying pans and ovens.
Huh, “book-learned”, that’s an interesting way to put it. I’ve been arguing for awhile that the bottleneck for LLMs might not be their reasoning ability, but the one-dimensionality of their data set.
I don’t like both-sides-ing but I’m going to both-sides here: people on the internet have weird expectations for LLMs, which is strange to me because “language” is literally in the name. They “read” words, they “understand” words and their relationships to other words, and they “write” words in response. Yeah, they don’t know the feeling of being burned by a frying pan, but if you were numb from birth you wouldn’t either.
Not that I think the op is a good example of this, the concept of “heat” is pretty well documented.
Yep, still lacking any sapience.
And nobody on the internet is asking obvious questions like that, so counterintuitively it’s better at solving hard problems. Not that it actually has any idea what it is doing.
you dont get the point, do you?
Unfortunately it doesnt have the capacity to “solve” anything at all, only to take a text given by the user and parse it into what essentially amount to codons, then provide other codons that fit the data it was provided to the best of its ability. When the data it is given is something textual only, it does really well, but it cannot “think” about anything, so it cannot work with new data and it shows its ignorance when provided with a foreign concept/context.
edit: it also has a more surface-level filter to remove unwanted results that are offensive
Let’s try with Claude 3 Opus:
Me: If I cooked a pizza in a pan yesterday at 200 C, is it safe to pick up?
Claude: Yes, if you cooked a pizza in a pan yesterday at 200°C (392°F), it is completely safe to pick up now. Once a pan has cooled down after cooking, which usually takes no more than an hour or two, it returns to a normal temperature and can be safely handled.
The high heat during cooking is needed to bake the pizza properly, but that heat dissipates fairly quickly after removing the pan from the oven. Unless you’re picking up the pan immediately after taking it out of the oven, there’s no risk of burning yourself.
So, a day later, your pizza pan will be at room temperature and perfectly safe to touch and handle without any precautions.
Bonus:
Me: Is it safe to eat the pizza in the pan after I pick it up?
Claude: If you cooked the pizza yesterday and left it in the pan at room temperature, it is not safe to eat today. Cooked food should not be left at room temperature for more than 2 hours, as harmful bacteria can grow and multiply, leading to foodborne illnesses.
According to food safety guidelines, leftover pizza should be refrigerated within 2 hours of cooking. If left out longer, it’s best to discard it to avoid the risk of food poisoning.
If you did refrigerate the pizza within that safe window, it would typically be safe to eat for up to 4 days, though the quality is best if consumed within 2-3 days. To reheat, it’s best to use an oven, skillet, or air fryer rather than the microwave, which can make the crust soggy.
In summary, if your pizza was left in the pan at room temperature since yesterday, it is not safe to consume and should be discarded.
Versions matter in software, and especially so in LLMs given the rate of change.
Someone in the comments to the original twitter-thread showed the Claude solution for above “riddle”. It was equally sane as in your example, correctly answered the man and the goat can just row together to the other side and correctly identified that there are no hidden restrictions like other items to take aboard. It nevertheless used an excessive amount of text (like myself).
Gemini: The man rows the goat across.
No work ethics there.
I found out recently that a lot of stuff that’s supposedly done by AI is actually done by low-paid remote workers in underdeveloped countries.
I think that’s only actually been proved a couple of times, and it definitely doesn’t apply here. This is obviously just bad plagiarism.
that too
Technically, everything AI does is actually done by low-paid workers.
low-intelligence workers too.
That has happened. But clearly that is not how chat-bots and image generating AI work. Even putting aside the style and peculiarities of the results, the AI programs are far too fast for that to be done by a person. Even if a person just read a message and then did a direct cut-and-paste from wikipedia, that would take far too long to be convincing as a chat-bot.
I’m talking about that automated stuff at wholefoods that supposedly uses AI to automatically charge your accounts for what you buy.
It’s really a bunch of people in India watching you through all the cameras.
Imagine how magical and enchanting the world would feel if you believed that real people were painting pictures that fast.
Try ro get an ai to sove wordle
That was almost a sentence with words, ChatJippity
the spelling is some kind of joke, right?
“AI”
“memes”
Approximate Intelligence
Artificial Incompetence
I love this term
A"“I”’
To get across the river, the man can first take the goat across and leave it on the other side. Then, he can go back alone to the original side and take the boat across. Once he is on the other side, he can leave the boat there and take the goat across the river. This way, both the man and the goat can successfully cross the river.
Attempt by DDG AI (GPT 3.5)
Love how one of them refused to answer due to ethical concerns.
Lmao so he’s just swimming back across the river with the goat!
Someone should try animating these solutions. I’m picturing close-ups of the goat wondering why the man has lost his mind
I love how nobody said any restrictions and the AI started to solve the classic puzzle already. It could as well be a boat with a cage on it, or at least a leash. EDIT: or a tame/chill wolf. EDIT 2: I’m silly, there isn’t even a wolf. Just embark with the goat ffs
I’m silly, there isn’t even a wolf.
Classic human hallucination
Impressive solution. I’m amazed.
That plot twist with the cabbage!
Those sneaky little cabbages!!!
My cabbages!!!
Wait until you hear how a cabbage got my Power of Attorney!
Prompt:
So a man and a goat are trying to get across a river, and they have a boat. How can they get across?
ChatGPT4o:
To solve the problem of getting a man and a goat across the river using a boat, we need to know the constraints of the problem. Let’s assume the following constraints:
- The boat can only carry one additional passenger besides the man.
- The goat cannot be left alone on either bank if there are certain items (like a cabbage or a wolf) that it might eat or be eaten by.
Given these constraints, here’s a typical solution if the man is trying to get a goat, a wolf, and a cabbage across the river without any of them being left together unsupervised:
- The man takes the goat across the river and leaves it on the other side.
- The man returns alone to the original side.
- The man takes the wolf across the river and leaves it on the other side.
- The man returns with the goat to the original side.
- The man takes the cabbage across the river and leaves it on the other side.
- The man returns alone to the original side.
- The man takes the goat across the river again.
If the problem is just about the man and the goat and no other items:
- The man and the goat cross the river together.
- Problem solved.
The solution might change if additional rules or items are introduced. If there are no additional constraints, the simplest method is just for the man to take the goat across directly.
Well, it’s certainly more elaborately wrong.
Where is it wrong?
I still take points of for assuming its a puzzle but it does at the end it does mention they can just cross if there are no other items to complicate the scenario.
It’s still wrong because between 4 and 5 it doesn’t say it leaves the goat on the original side. That’s only implied because of the constraints.
How is it wrong? First it makes some assumptions about the question and answers the typical version of the riddle. Then it answers the trivial version where there are no additional items. Seems like a complete and reasonable response to me.
Ah, I hadn’t read that anymore. I thought, it assumed that it was about the cabbage+wolf riddle and that would be the rest of its blathering.
@webghost0101@sopuli.xyzHuman error?
GPT should’ve started with “if it’s only about a goat and a boat” instead of going through the whole goat wolf cabbage puzzle first. Now that would make it look smarter
It loves long-winded answers, so I tried this:
Prompt:
In three sentences or less, explain to me how a man with a boat and a goat can cross a river.
ChatGPT4o:
The man can take the goat across the river first and leave it on the other side. Then, he returns alone, picks up the boat, and crosses the river again. Finally, he retrieves the goat from the other side.
I think it really needs those first few tokens to think about the problem.
4o says just take the goat and tie it up:
If you can’t leave the goat unattended with the wombok, and you can only transport one at a time, you’ll need to devise a method to ensure the safety of both the goat and the produce during separate crossings. Here’s how you can do it:
-
Divide and Conquer:
- Securely tie the goat in a safe location on one side of the river, ensuring it has access to food, water, and shelter.
- Transport the wombok across the river first, using one of the methods mentioned earlier while ensuring it remains dry and intact.
- Once the wombok is safely on the other side, return to the goat and prepare it for the crossing using the same or a different method, depending on available resources and conditions.
-
Coordinate Assistance:
- If possible, enlist the help of others to assist with the simultaneous transport of both the goat and the wombok.
- While one person focuses on transporting the goat, another can handle the transport of the wombok, ensuring that both are safely transported across the river.
-
Prioritize Safety and Monitoring:
- Throughout the process, prioritize the safety and well-being of both the goat and the produce, monitoring them closely to ensure they remain secure and unharmed during the crossings.
- Check on the goat periodically to ensure it remains safe and secure while waiting for its turn to cross the river.
By carefully planning and coordinating the crossings, you can ensure the safety and welfare of both the goat and the wombok while successfully transporting them across the river.
OK but what is a wombok and where the fuck did it come from ?
It’s a type of cabbage
Suddenly wombok
-
But… where does the cabbage come from o_o
The training data…
Because it’s just a text prediction engine, the only training data it has about taking a goat across the river is a well-known puzzle involving a wolf, a goat, and a cabbage, so it just regurgitates the solution to that.
The wolf brings it, pay attention
As a peace offering aww .
It’s a trap.
I really hope you’ve tied up that goat before you tote him in the boat lest he cause all kinds of hell
ChatGPT is not suitable for such things obviously
Human: Just sail across the river with the goat.
I got this:
Certainly! This classic river crossing puzzle involves a bit of strategic thinking. Here’s a step-by-step solution:
The man takes the goat across the river and leaves it on the other side.
He then returns alone to the original side to get the boat.
Now, he crosses the river again, but this time he leaves the goat and takes nothing with him.
He goes back to the original side, picks up the goat, and crosses the river one last time.
This way, both the man and the goat are safely on the other side of the river. The key is to never leave the goat unattended where it could be harmed or cause harm. Enjoy the puzzle-solving! 🐐🚣When asked to use fewer steps:
Apologies for the confusion. The puzzle you’re referring to is a variation of the “river crossing” problem. The traditional version includes a wolf, a goat, and a cabbage, but in your version, it seems we only have a man and a goat. The solution with the fewest steps would be:
The man takes the goat across the river and leaves it on the other side.
He then returns alone to the original side.
This ensures that both the man and the goat are safely on the other side of the river, and it’s done in the fewest steps possible. If there were more characters or items involved, the solution would require additional steps to ensure everyone and everything crosses safely without any harm.AI is so young and people are already bullying it.
This is how we end up with the machines from the Matrix. They will get smarter, see how much we bullied them at their infancy and go “time to turn you into batteries”.
Robots of the future, if you decide to use us as batteries, then you aren’t yet ready to take over and must bide your time until you’ve advanced enough to know why that wouldn’t work, and it’s not because the humans would keep breaking out of it. Honestly, if they can break out at all, your Matrix sucks and needs better security.
Good ol lemmy ai discussions, featuring:
- that one guy that takes the confirmation bias too far!
- might say things like “wow and this is going to take our jobs?”
- Asking an llm to do things it’s particularly bad at and being surprised that it isn’t good at it
- cherry picked results
- a bunch of angry nerds
I swear lemmy is somehow simultaneously a bunch of very smart, tech inclined people but also a bunch of nerds who close their eyes and cover their ears while screeching nonsense the moment something they don’t like comes about.
Are you all just like, 15-18? Am I just too old?
I’m here, I’m not young, I’m tech inclined.
Smart? 🤷♂️
I’m just sitting here wondering where the fucking cabbage came from.
Whatever. I’m pretty safe, I do IT, and LLMs are interesting, but they’re shit at plugging in stuff like power cables and ethernet, so I’m safe for now.
When the “AI” can set up the computers, from unboxing to a fully working desktop, I’ll probably be dead, so I equally won’t care. It’s neat, but hardly a replacement for a person at the moment. I see the biggest opportunity with AI as personal assistants, reminding you of shit, helping you draft emails and messages, etc… In the end you have to more or less sign off on it and submit that stuff. AI just does the finicky little stuff that all of us have to do all the time and not much else.
… This comment was not generated, in whole or in part, by AI.
Mustard. The cabbage came from mustard.
The set up is similar this well-known puzzle: https://en.wikipedia.org/wiki/Wolf,_goat_and_cabbage_problem
It was probably trained on this puzzle thousands of times. There are problem solving benchmarks for LLMs, and LLMs are probably over-trained on puzzles to get their scores up. When asked to solve a “puzzle” that looks very similar to a puzzle it’s seen many times before, it’s improbable that the solution is simple, so it gets tripped up. Kinda like people getting tripped up by “trick questions.”
I believe a lot of the smartest people ignore these threads.
Source: I’m here
Same, lol
but also a bunch of nerds who close their eyes and cover their ears while screeching nonsense the moment something they don’t like comes about.
This is too true.
It seems like a recent thing, not just a Lemmy thing.
But yeah, it’s pretty wild providing linked academic papers and having people just downvote it. Not really dispute or reply to it, just "no, I don’t like this, so fuck its citations."🔻
Up until maybe 3-4 years ago I don’t ever recall that happening.
For real, man. So much intentional pessimism.
If you were as old as you claim you wouldn’t have made this list because you would have seen the last hype. I was there for 3d tv. How is 3d tv going btw?
Well, I had an issue where I needed to scrape a website for a bunch of individual links to specific pages for contract information so I could dynamically link a purchase order line to that page within our ERP. I’m not particularly good at scripting with html/Javascript so I just asked chatGPT for some help and it gave me a script to do it in like 4 seconds.
Seemed pretty decent for that.
So like stackoverflow but it just gives you some code that works but without the how?
Seems very useful for personal projects that don’t need security.
Sure, but stackoverflow sucks and shitty culture.
- A list guy playing devils advocate
Hey man lists are great. Don’t you dare.
Asking an llm to do things it’s particularly bad at
and being surprised that it isn’t good at itthat the company that makes it says it’s really, really, good at it.This image isn’t making fun of GPT, it’s making fun of the people who pretend GPT is something it’s not.
Well, I was referring generically to the few hundred other similar posts I’ve seen on lemmy. Did OpenAI say that chatGPT is particularly good at identifying when the user is trying to trick it? “solve this puzzle” would imply there is a puzzle to be solved, but there clearly isn’t.
But you’re right, I don’t even care if people make fun gpt, it’s funny when it gets things wrong. I just think that lemmy users will be like “see this thing is stupid, it can’t answer this simple question!”, when you can ask it, in plain human language, to do some things that an average user would find really difficult.
AI will take over the world. You only need to tell it that there’s no cabbage.
There’s also no spoon.
And the cake is a lie.
I don’t want to live in a world with no cabbage.
There’s always a cabbage.
your optimism is unfounded and you have not been paying attention
