Buyee seem to love nickel and diming on package inspection fees and packing materials when consolidating (they're presumably already safely packed by the seller? just put it in a box for me), then supposedly have in some cases opened sealed items etc. Then even though they charged for the individual items on checkout, they seemed to bill for Japanese domestic shipping for everything only once you picked shipping for the final package? Very confusing and hard to keep track of.

In the same fashion, LLMs have to pay for themselves to keep the trendlines going. In a whole-systems -sense, mind, not "$2000/month is cheaper than hiring a developer" while the rest of the economy collapses.

Or indeed with allies, as Europe is just finding out...

Have Anthropic actually said anything about the amount of false positives Mythos turned up?

FWIW, I saw some talk on Xitter (so grain of salt) about people replicating their result with other (public) SotA models, but each turned up only a subset of the ones Mythos found. I'd say that sounds plausible from the perspective of Mythos being an incremental (though an unusually large increment perhaps) improvement over previous models, but one that also brings with it a correspondingly significant increase in complexity.

So the angle they choose to use for presenting it and the subsequent buzz is at least part hype -- saying "it's too powerful to release publicly" sounds a lot cooler than "it costs $20000 to run over your codebase, so we're going to offer this directly to enterprise customers (and a few token open source projects for marketing)". Keep in mind that the examples in Nicholas Carlini's presentation were using Opus, so security is clearly something they've been working on for a while (as they should, because it's a huge risk). They didn't just suddenly find themselves having accidentally created a super hacker.

"Danny leaves the scene" (because it's just a bunch of kids with scanners and he's got a job at Eidos now) never forget!

Check https://m4de.com/?tag=archives

There's some semi-apologetic interest in ML, esp. smaller local models, in the "permacomputing" (don't like the term but whatev) sphere. But I don't know if there's much of a conversation around LLMs. With all the hype and how resource intensive and externalities-heavy they are, I can see wanting to draw a line, but it's sad to see it become a purity test.

Lately the discussion around this has had me thinking of the William Köttke quote "not only is it ethical to use the resources of the current system construct the next one, ideally, all the resources of the current system would be used to that end".

I think that if the situation was as dire as it's made out to be (I think it is) and projects like uxn were a serious attempt at a mitigating response (less convinced, as cool as they are), there's room for a conversation about beneficial-detrimental (rather than good-evil). Then we could discuss whether it's a good idea to use LLM-based tools when they are available to help build out infrastructure that runs without them, whether there's a nuance as to at what level of automation we draw the line (Ivan Illich, tools vs machines etc), human augmentation vs replacement, the cognitive load stuff Keeter's post touches on and so on.

Unfortunately, part of the polycrisis seems to be a difficulty in discussing things clearly.

> a bit tone deaf

Agree