It's not novel at all, we do it all the time. It's very common to say "program X did Y" without making the conversation about blame or responsibility. But when the program is an AI agent suddenly using it as a subject of a sentence and saying that "agents did X" becomes a sensitive topic for some people.
And we surely need this for AI as the swiss cheese zone is getting rather huge.
My position, and the position of a large number in AI safety, is that you cannot build an intelligence that is both general and safe. The closer you get to generalized the more options the system has to do things that are wildly unsafe beyond human imagination.
This puts the AI labs in a serious bind while holding a bag filled with billions of dollars of debt.
Worse this puts governments in a multi-polar problem where even if the big public labs get shut down, black budget operations have a lot of free reign to make agentic digital weapons. Governments are not well known to take a lot of responsibility when their weapons cause damage unless they lose.
You could've included a little more nuance. An interpretation of your response is "being the most popular remotely accessible software" != "the most attacked", which would need defending.
> as purely empirical as the gradient descent loops
Are you suggesting that gradient descent is an empirically found and not understood technique? It was originally proposed by Cauchy in 1847, its properties are very well understood.
You might be referring to properties of the domains its being applied to.
I’m not saying gradient descent was empirically discovered, I’m saying that its use in machine learning (or elsewhere, I suppose) is itself a form of empiricism in that what it is is essentially a repeated observe/measure error/adjust cycle.
> I’m not saying gradient descent was empirically discovered, I’m saying that its use in machine learning is itself a form of empiricism. A repeated observe/adjust-based-on-data cycle
The data is the input, the output is to generally find the lowest amount of a loss function. It’s a greedy approach because brute forcing is inefficient.
It’s no more empirical than a greedy algorithm for scheduling.
> It’s no more empirical than a greedy algorithm for scheduling.
Right, GP is drawing a distinction between search, ie mechanical exploration of a space, with understanding, ie having a map of the territory such that you don’t need trial and error.
> its use in machine learning (or elsewhere, I suppose) is itself a form of empiricism in that what it is is essentially a repeated observe/measure error/adjust cycle.
"Empiricism" implies that the technique is based on observable, but not mathematically proven foundations. If a problem space is convex, gradient descent is guaranteed to converge to a global optimal solution, regardless of whether you know the exact formulation of the space.
Applying it when you don't understand if a space is convex is another question, but that's not a fault of gradient descent.
I was so primed to expect crazy I had to reread the article and your comment to notice "this bit of English isn't that crazy".
For those who won't read the article, either the word "a" or the word "an" is used before a word. Which is used depends on the first sound of the next word (not whether the next word starts with a vowel, as is commonly understood).
With this rule, only "129 of the 32,455 words in my list needed exceptions.", which is surprisingly regular.
I'm on my phone so maybe I missed something but I didn't even see that this was about exceptions to the rule? It seemed to be more about cases where the first sound couldn't be recovered orthographically. He special cased herb, for instance, because Americans inexplicably pronounce it "erb", or one and onerous because one is w and the other is on.
I can't think of a single true example to the real rule, it's an actually extraordinarily strong and simple rule as English goes.
That's how I've always understood the rule as a native speaker, even though I don't think it was ever taught (to me) in school growing up. It all depends on how the next word sounds.
You're specifically frustrated about unbounded stack recursion exhausting the stack, triggered by multiple thousands of directories? It doesn't sound like this is about multi-thousand-deep directory structures, it sounds like it's about something else.
Because even diving into it, I would agree with a prioritisation decision that puts this bug down the bottom of a priority list.
Canonical is just rushing the switch because they want to get rid of software with GPLv3 license, not because there is any technical merit for doing so.
If they want the default to be non-GPLv3 Rust-based that's fine, but some of us don't care (and want the same behaviour everywhere, like on RH-based systems we may also have) and they should leave the GNU as an option. Potentially both could be installed at the same time (it's what update-alternatives is for after all).
That's a better thing to discuss, and I'd expect there to be better bugs to talk about to go with it. Stack exhaustion in an unrealistic environment should be fixed, but a "stop everything" bug it ain't.
But can they compete with my shade grown software? It does result in a 40% markup, but there's no putting a price on being raised in a loving environment, is there.
The comment concludes with "Conclusion: no, bad advice, except for very simplistic scenarios.", so I don't think I can agree with you on this. My point was that these facts don't (in my analysis) support the conclusion, and suggest a contradictory stance (in particular: planning in advance to dual-boot multiple separate distros, but not being willing to adjust the configuration after installation).
reply