At least some of that are people answering wrong on purpose to be funny, contrarian, or just to try to hurt the study.

This is necessary for sounding like reasonable language and an inherent reason for "hallucinations". If it didn't have variation it would inevitably output the same answer to any input.

I'm sure AI will do those things at some point. Nobody expected the same of our microorganism ancestors.

In fact it cannot ever be a useful way of extracting information. Even just randomly guessing is a better way to get the information you want than torture.

Our microorganism ancestors also did all those things, and they were far beyond anything an LLM can do. Turning a given list of words into numbers, doing a string of math to those numbers, and turning the resulting numbers back into words is not consciousness or wisdom and never will be.

Yeah i remember that Ed article ! I don't think the technical aspects are relevant to the newer generation of models, but yeah of course any attempt to compress inference costs can have side effects : either response quality will degrade for using dumber models, or you'll have re-inference costs when the dumb model shits its pants. In fact the re-inference can become super costly as dumber models tend to get lost in reasoning loops more easily.

It's also the case that people are mostly consistent.

Take a question like "how long would it take to drive from here to [nearby city]". You'd expect that someone's answer to that question would be pretty consistent day-to-day. If you asked someone else, you might get a different answer, but you'd also expect that answer to be pretty consistent. If you asked someone that same question a week later and got a very different answer, you'd strongly suspect that they were making the answer up on the spot but pretending to know so they didn't look stupid or something.

Part of what bothers me about LLMs is that they give that same sense of bullshitting answers while trying to cover that they don't know. You know that if you ask the question again, or phrase it slightly differently, you might get a completely different answer.

Even if you retooled the LLM to not randomize the output it generates, it can still create contradictory outputs based on a slightly reworded question. I'm talking about a misspelling, different punctuation, things that simply wouldn't cause a person to change their answer.

(And that's assuming the LLM just got started from scratch. If you had any previous conversation with it, it could have influenced the output as well. It's such a mess.)

3 in 10 people get this wrong‽‽

Maybe they're picturing filling up a bucket and bringing it back to the car? Or dropping off keys to the car at the car wash?

It's not literally guessing, because guessing implies it understands there's a question and is trying to answer that question. It's not even doing that. It's just generating words that you could expect to find nearby.

I'm pretty sure Google's AI is fed by the same spider that goes out and finds every new or changed web page (or a variant of that).

As soon as someone writes an article about how AI gets something wrong and provides a solution, that solution is now in the AI's training data.

OTOH, that means it's probably also ingesting a lot of AI generated slop, which causes its own set of problems.

I like the idea of referring to LLMs as “spicy autocomplete”.

Humans are running out of "tricky" puzzles to retreat to.

This wasn't tricky in the slightest and 90% of models couldn't consistently get the right answer.

That's why when I need help with something I don't go out and ask a random human.

That 71.5% is still a higher success rate than 48 out of 53 models tested. Only the five 10/10 models and the two 8/10 models outperform the average human. Everything below GPT-5 performs worse than 10,000 people given two buttons and no time to think.

It's tricky in the sense that it requires abstract reasoning.

What open weight model do you think is best right now?

Dirtying the car on the way there?

The car you're planning on cleaning at the car wash?

Like, an AI not understanding the difference between walking and driving almost makes sense. This, though, seems like such a weird logical break that I feel like it shouldn't be possible.

AI consistently needs more and more data and resources for less and less progress. Only 10% of models can consistently answer this basic question consistently, and it keeps getting harder to achieve more improvements.

Ha ha ! Nice one, the paid version is worse

You think microorganisms can reason? Wow, AI haters are grasping for straws.

Honestly, I don't understand Lemmy scoffing at AI and thinking the current iteration is all it ever will be. I'm sure some thought that the automobile technology would not go anywhere simply because the first model was running at 3mph. These things always takes time.

To be clear, I'm not endorsing AI, but I think there is a huge potential in years to come, for better or worse. And it is especially important to never underestimate something, especially by AI haters, because of what destructive potential AI has.

Why would the car already be at the car wash if you ask it wether or not you should drive there?

Having read the article (you read the article right?) what gave you the impression the AI was asked the question at different points in time?

You're assuming AI "think" "logically".

Well, maybe you aren't, but the AI companies sure hope we do

I didn't get it right until people started taking about it.

Hey LLM, if I have a 16 ounce cup with 10oz of water in it and I add 10 more ounces, how much water is in the cup?

Why wouldn't it be? How often have you thought, I wonder if I should drive my car to the carwash, maybe I should ask someone?

That's the thing: it is a nonsensical question, the only sense of it is if YOU need to get where the carwash and car is because you must be asking about something else.

I am not saying AI is making any sense, it cant. But if you follow the weights and statistics towards the solution for this question, it is about something else other than driving the car to the car wash, because nothing in the training would have ever spelled that out.

Well I did interview at Microsoft once a long time ago. They did ask some stupid questions... lol

Absolutely not, I'm still just scratching my head at how something like this is allowed to happen.

Has any human ever said that they're worried about their car getting dirtied on the way to the carwash? Maybe I could see someone arguing against getting a carwash, citing it getting dirty on the way home — but on the way there?

Like you would think it wouldn't have the basis to even put those words together that way — should I see this as a hallucination?

Granted, I would never ask an AI a question like this — it seems very far outside of potential use cases for it (for me).

Edit: oh, I guess it could have been said by a person in a sarcastic sense

LOL! That is a great answer.

I have a Microsoft story. I know some one who was hired to stop them from continuing an open source project. They gave them a good salary, stock options, and an office with a fully stocked bar. They said do whatever you want, they figured they would get a good developer and kill the open source competition (back in the Ballmer days).

Sadly, given money, no real ambition to create closed source software, they mostly spent their days in their office and basically drank themselves to death.

Microsoft just kills everything it touches.

This is why computers are expensive.

I think you misread it - 71% said drive. 29% is still pretty bad, but it is kind of a "who is buried in grants tomb" question.

The question is based on assumptions. That takes advanced reading skills. I’m surprised it was 71% passing, to be honest. (The humans, that is)

This is why it’s a bad question to test a computer with.

It requires understanding what is being asked.

The straw I'm grasping at in this example is a reasonably well-accepted scientific consensus, but you do you.

Turning a given list of words into numbers, doing a string of math to those numbers, and turning the resulting numbers back into words is not consciousness or wisdom and never will be.

Neither is moving electrolytes around fat barriers.

Understanding the intent of the question *and understanding why it could be interpreted differently *\and understanding why is it is a poorly phrased question are not related to autism. (In my case)

I want to wash my car. No location or method is specified. No 'at the car wash'. No 'take my car to the car wash' . No 'take the car through the car wash'

A car wash is this far. Is this an option? A question. A suggestion. A demand?

Should I walk or drive? To do what? Wash the car? Ok. If the car wash is an option, that seems very far. But walking there seems silly. Since no method or location for washing the car was mentioned I could wash my own car.

Do you see how this works?

Yes, you can infer what was implied, but the question itself offers no certainty that what you infer is what it is actually implying.

Yes. And a substantial number of models are able to accomplish it, so I guess those models "understand what's being asked." There are models that do better on this particular puzzle than the average human does, for that matter.

I started experimenting with the spice the past week. Went ahead and tried to vibe code a small toy project in C++. It’s weird. I’ve got some experience teaching programming, this is exactly like teaching beginners - except that the syntax is almost flawless and it writes fast. The reasoning and design capabilities on the other hand - ”like a child” is actually an apt description.

I don’t really know what to think yet. The ability to automate refactoring across a project in a more ”free” way than an IDE is kinda nice. While I enjoy programming, data structures and algorithms, I kinda get bored at the ”write code”-part, so really spicy autocomplete is getting me far more progress than usual for my hobby projects so far.

On the other hand, holy spaghetti monster, the code you get if you let it run free. All the people prompting based on what feature they want the thing to add will create absolutely horrible piles of garbage. On the other hand, if I prompt with a decent specification of the code I want, I get code somewhat close to what I want, and given an iteration or two I’m usually fairly happy. I think I can get used to the spicy autocomplete.

Ai is not human. It does not think like humans and does not experience the world like humans. It is an alien from another dimension that learned our language by looking at text/books, not reading them.

Understanding the intent of the question *and understanding why it could be interpreted differently *\and understanding why is it is a poorly phrased question:

There are 3 sentences.

I want to wash my car. No location or method is specified. No 'at the car wash'. No 'take my car to the car wash' . No 'take the car through the car wash'

A car wash is this far. Is this an option? A question. A suggestion. A demand?

Should I walk or drive? To do what? Wash the car? Ok. If the car wash is an option, that seems very far. But walking there seems silly. Since no method or location for washing the car was mentioned I could wash my own car.

Do you see how this works?

Yes, you can infer what was implied, but the question itself offers no certainty that what you infer is what it is actually implying.

The AI was asked the same question repeatedly and gave different answers, due to its randomised structure.

People will also often do this (I have, personally), but because our actions seem to be strongly influenced by time-dependent stuff (like sense perception and short-term memory contents), I'd expect you'd need to ask at different times.

LLMs can't learn. It's one of their inherent properties that they are literally incapable of learning. You can train a new model, but you can't teach new things to an already trained one. All you can do is adjust its behavior a little bit. That creates an extremely expensive cycle where you just have to spend insane amounts of energy to keep training better models over and over and over again. And the wall of diminishing returns on that has already been smashed into. That, and the fact that they simply don't have concepts like logic and reasoning and knowing, puts a rather hard limit on their potential. It's gonna take several sizeable breakthroughs to make LLMs noticeably better than they are now.

There might be another kind of AI that solves those problems inherent to LLMs, but at present that is pure sci-fi.

It's not just a copy machine, it learns patterns...without knowing why the fuck.

Look, human conversations are full of context deduction and inference. In this case "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?" states my random desire, a possible solution and the question all in one context. None of these sentences make sense in isolation as you point out, but within the same frame they absolutely give you everything you need to answer the question of find alternatives if needed.

Sorry for the random online stranger diagnosis but this is just such an excelent example of neurodivergent need for extreme clarity I couldn't help myself.

I agree that it should be able to infer the intent, but I stand by that it remain somewhat unclear and open to interpretation. Eg, If such language was used in a legal contract, it would not be enough to simply say, well, they should understand what I meant.

The people doing this test, I'm sure, are not linguistic masters, nor legal scholars.

There are lines of work where clarity is essential.

And what if my question actually was asking, should I just go for a walk instead of driving that far?

I know the answer. But as 30% demonstrated, clarity IS needed.

He's not totally serious he's cardfire. Silly human

Can you explain how quorom sensing is reasoning and exercising logic?

It's dumber than that actually. LLMs are the auto complete on your cellphone keyboard but on steroids. It's literally a model that predicts what word should go next with zero actual understanding of the words in their contextual meaning.