Of course you could also argue that human intention comes from largely deterministic processes emerging from the brain. That may eventually perhaps lead to all figures of speech involving things like intentionality meaningless.
This type of response is just stochastic parrotry, rather than displaying evidence of actual <whatever cognitive trait we're overconfidently insisting LLMs don't have>.
Yet more evidence that LLMs are more similar to humans than we give them credit for.
Never stops fascinating me how folks are arguing this kind of thing. Why make up an explanation for why this obvious mistake is actually some kind of elaborate 4D chess sarcastic "intention"? It's a simple machine, its network just didn't support making up a new Toy Story character. That's it! Simple as that! Occam's Razor anybody?
Or yes, maybe the regex I wrote the other day which also had a bug that missed replacing certain parts also had an "intention". It just wanted to demonstrate how fallible I am as a human, so it played this elaborate prank on me. /s
...Because Occam's razor is not assuming it's a "mistake"?
There's a thread full of people saying how clever humorous they find almost every headline.
The real 4D chess is dogmatically assuming it is not assuming it managed to by pure accident succeed in that dozens of separate times, because your dogma refuses to incorporate evidence to the contrary.
Occam's razor is that this system which no one actually understands the emergent capabilities of, and is convincing so many people it has intention... has intention.
There is no language that makes it impossible to have any kind of bug ever. The safety languages like Rust offer is around memory, not bad configuration or faulty business logic.
If they have a HTTP API using standard authentication methods it's not that difficult to create a simple wrapper. Granted a bit more work if you want to do things like input/output validation too, but there's a trade-off between ownership there and avoiding these kinds of supply-chain attacks.
If you aim for 100% coverage of the API you're integrating with, sure. But for most applications you're going to only be touching a small surface area, so you can validate paths you know you'll hit. Most of the time you probably don't need 100% parity, you need Just Enough for your use-case.
To my understanding, there's less surface area for problems if I have a wrapper over the one or two endpoints some API provides, which I've written and maintain myself, over importing some library that wraps all 100 endpoints the API provides, but which is too large for me to fully audit.
Node itself is still fine and you can do a lot these days without needing tons of library. No need for axios when we have fetch, there's a built-in test runner and assertion library.
There are some things that kind of suck (working with time - will be fixed by the Temporal API eventually), but you can get a lot done without needing lots of dependencies.
I used to work for Suncorp (~10 years ago) in Brisbane square, pretty sure they are no longer using the building and it’s now a super company.
I’ve not watched Bluey (don't have kids of my own or niblings the right age), but am looking forward to when the time comes, so I can point out places I know.
Like many here, I grew up in Brisbane but moved elsewhere for opportunities etc. Given the cost of everything there these days I doubt I’ll make a permanent move back, but I do get nostalgic seeing the locations and the little details represented so well.
