[coldtea]

Can a 5 year old write a substantial program on spec, that passes the requirements and given tests, in a few minutes?

If not, then perhaps this comparison is not the be all end all.

"A ship is useless, it can't drive over land..."

[dominotw]

5 year olds and ai both have jagged intelligence.

also its AI not "artificial code generation intelligence" . Ship is your view of the product to shoehorn into something specific.

[JBits]

But it demonstrates that LLMs struggle with basic reasoning. A criticism of LLMs is that they're imitating without a understanding of what they're doing and without a clear plan, so this inability to solve a simple logic puzzle is very relevant. If LLMs didn't struggle with reasoning problems then something like ARC-AGI wouldn't exist.

[Eisenstein]

Its a question designed to fool the AI. It's like saying that a person doesn't understand the limitations of reality when they fall for a magic trick.

[watutalkinbout]

These aren't the same thing.

You can't fool an AI because it isn't using its own judgement.

[Eisenstein]

You can fool an AI as evidenced by them being fooled. They demonstrably appear to be working through a problem and get fooled by the wording of the problem. If you think differently, merely asserting it is not the way to convince people that what they see is wrong.

[watutalkinbout]

You can't fool an inanimate object.

An LLM is (effectively) just a really, really elaborate "choose your own adventure" book.

It's not "working through" problems, it's just tracing a route through an pre-defined information space. It's not actually thinking, it just does a good impression of it.

[coldtea]

>You can't fool an inanimate object

You can't constrain foolability to just the animate objects.

Foolability is based on intelligent processing.

To what degree such an intelligence needs to be developed is not defined at all (we can, for example, trivially fool a doog using one of many tricks, and a dog is hardly general intelligence).

Thus far, the inteligent processing we knew concerned animate objects. But we now have developed software/hw combos that exhibit intelligent processing. Is it enough to actually be intelligent or to be fooled?

We don't know. We do know it's enough to appear intelligence, convince people that it is intelligent, and to appear to be fooled.

But even if we think of LLM AI as mere mechanistic process with no emergent intelligence, who said one can not fool a mechanistic process? We can fool even simple pre-LLM gaming AI systems (based on simplistic heuristics) just fine.

[Eisenstein]

Please explain why the mechanism of the LLM generating output precludes it from being able to be fooled without using use tautologies or reducing to substrate for explanation.

[watutalkinbout]

You're describing a tool.

Tools can do very useful things, but they aren't intelligent.