NateEag

You want a VM, not a container, because VMs are harder to escape.

since my employer is compelling me to throw out my expertise and rot my brain with Claude, I've been using clawk to sandbox instances in VMs: https://github.com/clawkwork/clawk

It's not perfect, but it works pretty well, and has done a nice job of forcing me to think through exactly what I'm allowing Claude access to and why, and what the risks are.

Yes - a deterministic tool is just fine for verifying that a template has been followed.

It's not quite as simple as "check for one line," but it's not hard to have a useful check that costs far fewer resources to run than an agent would.

You don't need an agent to check if the PR matches the template.

Sure, as one raised by an ER doctor, I have had a very up-close-and-personal view of many doctors being arrogant.

I can't see any of the ones I've known not stepping aside on discovering they're spontaneously, unpredictably losing the memory of how to do surgical procedures.

I wouldn't go so far as to day no doctor would ever keep going on the face of that happening, but I will say to do that would be utterly inhumane.

> Somehow Opus 5 refusing after understanding its limitations felt like the most humane thing in the article.

This is an unfortunately true and insightful comment.

Some of us are still really sad, and have discovered we can't really make much more, because we're still trying to understand everything these bots we've been required to use vomit out.

That pretty much entirely negates any speed gains.

Thank you, much appreciated!

This is a very interesting datum.

Can you point me to a source?

> but this seems a long way from saying that logical fallacies are to rationalists as heresies are to (say) Christians.

Logical fallacies do nothing to show an argument is wrong.

All they do is demonstrate that a chain of reasoning cannot be a logical proof that the conclusion must be true, given its premises.

Dismissing arguments because "they're fallacious" is actually not so far removed from dismissing ideas because they're heretical.

In each case, what matters is whether the ideas are true, not whether your interlocutor proved rigorously it is true.

As one who has lurked in Rationalist corners of the web and learned useful things there without ever really buying in, I can confirm it really is treated rather like a holy text.

I've seen Rationalists defend it against very fair and accurate criticisms (see this whole thread for some of those), completely failing to recognize the complaints and objections are often legitimate.

I believe I've seen people cite it as the thing that brought them into the Rationalist community.

It's not uncommon to see people quote it in varying capacities - as a reminder of what "good rationality" looks like, as a cultural shibboleth, as a criticism of an idea they disagree with, and more. As an ex-evangelical who's read the Protestant Bible well over five times, this behavioral pattern looks very, very familiar.

I'm not going to spend the hours digging to find actual evidence I'm correct, and it's possible I'm being unfair and misremembering. I'm just a guy on the Internet - don't take this as anything other than "Here's my recollection. Take it for what it's worth."

The label of "Torah" did not strike me as wrong, though.

That link is unlikely to be an unbiased accounting.

Speaking of long-winded rationalist blog posts, the evidence for that claim is this long, detailed, full-of-receipts accounting of David Gerard's use of RationalWiki (though mostly Wikipedia) to push his personal agenda in subtle ways while scrupulously following the rules to avoid being caught at it:

https://www.tracingwoodgrains.com/p/reliable-sources-how-wik...

To be fair, the Rationalists _do_ accept the two-word rebuttal "Chesterton's Fence," in principle.

...though yes, in practice if your invoke or, they're fairly likely to explain in elaborate length why it doesn't apply here.

> `claude` can be considered a version of `rm` that can do what `rm` does (and other things - including things you don't want).

It can't, though.

Only when connected to a harness can Claude remove files from disk.

And, at that point, it's using existing libraries or tools to do it - `rm` if it uses bash, the `remove` stdio function if it uses C or a C-implemented programming language to write a custom tool for the job, or one of many other options.

It remains true, though, that Claude cannot itself remove files - just be your proxy to guess what files it's probable you were referring to, that you likely wanted to remove them, and then use another tool to actually do so.

As you note, it will be wrong about that sometimes, and it will likely be unpleasant if you haven't prepared for the eventuality.

> When people say "AI is just a tool", I think they mostly know what they mean and we all do, too.

When I hear that phrase, I think the speaker is either too naive to live, or is intentionally trying to sidestep anyone bringing up the manifold disaster-level issues with genAI systems.

> Notice your Webster’s doesn’t include the insult usage.

What exactly do you think

> a word of reproach

means?

I am a professional programmer whose employers have compelled me to use genAI.

I have no qualms about saying I hate it, every workday is utterly miserable, I can see my skills eroding, and I see almost nothing but downside to it.

FWIW, a past manager once dubbed me "the Luddite programmer," a badge I wore with pride.

Make no mistake, I do love computers, and writing software. Steve Jobs' metaphor of a "bicycle for the mind" captures well the role they played for me in my teens and twenties.

Alas, LLMs are like a freight train for shipping all minds to one destination, instead of freeing them to all find their individual ways to unique places.

...and also, the freight train's destination is hell.

NeXTSTEP. Grandparent is wrong.

Search this document for "Historical Note" for the canonical explanation:

https://developer.apple.com/library/archive/documentation/Co...

My apologies - I did not read the thread carefully enough.

Maybe, if you don't like how the company treats its customers and craps all over the commons, you could not use their products.

And, given the insanity that's been happening like the HuggingFace hack, maybe slower development would be a VERY GOOD THING.

No object, labor, or knowledge has inherent economic value.

Its value is subjective, determined by what the market (i.e., a bunch of humans) will pay for it.

Given the gigantic pile of money that's spent on music, film, and books every year, it's quite clear that there can be large amounts of economic value in a piece of art.

That modern technology makes it easier to take an artist's with with absolutely no recompense, and that many choose to do so regardless of the maker's wishes, does not change that.

It just underlined what's been obvious since the dawn of civilization, that many humans are happy to ignore what other humans want and to enrich themselves at the expense of others.

...which brings us back full circle to the actual article, which is a clear illustration of the OpenAI executive team's conscious choice to do exactly that.

"Word" there is Logos:

https://en.wikipedia.org/wiki/Logos

There's a lot more to it than just "language."

There's a crucial distinction between "can't sue you" and "would likely lose the case".

You need to have enough money to actually survive the lawsuit, even if it's a slam-dunk to win (and they almost never are).

I don't think that holds at all.

There are so many variables that go into upvoting and downvoting, and we can't even see the actual scores on HN posts.

The data you refer to isn't available, AFAIK, and even if it were we'd have to account for the hundreds of confounding variables that contribute to post scores being what they are.

For those of us who are aphantasiac, it is not possible to remember what we have seen ourselves.

The camera and sketchpad are the only options we have.

Yes, in a huge crowd of people, you will get differing opinions.

Do you have any other obvious facts you'd like to pretend are somehow relevant?

> humans take responsibility for the failure the same way they take responsibility of a photocopy is too dark.

You mean, they notice it's too dark right after making it, change the settings, do it again, and give you the good copy?

Because yes, that's my experience of humans.

For sure.

But they do learn and improve.

The models don't (yet).

And when they do, you can train them or fire them, and they learn not to do it.

LLMs change not a whit, and there's no one to take responsibility for the failure (and thus no way to fix it).

As the new variation on the old theme has it, "A computer can never be held accountable, and so very many people are trying to get them make management decisions."

> it's a lot more often targeted at Microsoft and Google software than at "vibe coded slop".

What makes you confident the Microsoft and Google software isn't vibecoded slop?

Certainly Microsoft has made a lot of noise about all their software being LLMgen these years, and also gotten into hot water with their users for Windows 11's crappiness.

I'm not a mathematician, nor do I know almost anything about the discipline, but it sounds like maybe you do.

What teams of people spent an entire career working together, focused entirely on Navier-Stokes or problems they suspected were related, without any "publish or perish" concerns?

I realize tenure is a thing, but my limited understanding is that a significant amount of time is spent earning it, once you account for Ph.D. program and the years of needed to be granted it.