sfink

6 ms·

31.
by sfink 1mo ago
I agree with the substance of this article, and disagree with the author's reaction to it. The part I agree with: It is true that watermarking can be done by "just" swapping one PRNG for another, and it is even true that with
32.
by sfink 1mo ago
Be grateful. I couldn't spot AI writing for a while, and I was suddenly enjoying reading clear and punchy text about all kinds of topics that people hadn't been talking much about. Then at some point it "clicked" and I w
33.
by sfink 1mo ago
Yeah, I remember reading about Anthropic's experiments with dropping out different things, and dropping the thinking is ok for compaction but pretty awful while it still fits in the context window. I could imagine doing it adaptively -
34.
by sfink 1mo ago
Then wake up, they're happening. Police stalking women with Flock camera data. Period/pregnancy data used by ex- and current boyfriends or law enforcement. Near Intelligence designed Planned Parenthood as a trackable brand, enabli
35.
by sfink 1mo ago
I'm going to push back on your pushback and say you're not disagreeing with me. I was referring to a hypothetical future where everything has gone through AI to some degree, and for the purpose of the argument it has been modifi
36.
by sfink 1mo ago
The author justifiably attacks the notion that "NP-hard" == "too hard to solve in practice", but then makes the opposite error: > Everyone knows you can tackle those with heuristics, but you don't have to sacrifi
37.
by sfink 1mo ago
I'll list some possible concerns. But first, can we agree that Mickey D's is probably not going to be a lot better at keeping this data secure than, say, my bank or government or insurance company? All of which have been hacked, r
38.
by sfink 2mo ago
Perhaps. But it seems like your beef is not with the presence of watermarking, it's with what people will use that watermarking for. You're not directly harmed by that blog post being labeled as AI generated. In a hypothetical (bu
39.
by sfink 2mo ago
I don't know how this watermarking works, but I don't need to in order to understand some things that a lot of this conversation seems to be missing. First, the article doesn't talk about adversarial usage. As in, it's n
40.
by sfink 2mo ago
Title has a weird spin. I'd say that "stronger" is more accurate than "older". Ironically, MV2 blockers can do dynamic updates and so are often "newer" in the relevant sense. I do wonder if we haven't
41.
by sfink 2mo ago
So: "welp, the horse is out of the barn. Time to close the door, I guess?"
42.
by sfink 2mo ago
The part about performance seems surprising to me. Sure, it's slower, but shouldn't the baseline be something like: first run the old borrow checker, and only run the new one if it finds a problem? Given that the checking is a fun
43.
by sfink 2mo ago
https://searchfox.org/firefox-main/rev/878b64a4c024f657de81f... but note line 111: > This section describes what to do if you're in a Jujutsu (jj) repository. If the user is not using jj, good luck and try
44.
by sfink 2mo ago
Heh, it's a fun reminder that different people work on different things. Sure, I could load up my build machine with mac/windows/linux/android, with multiple versions of each. Emulating x86+x86_64+arm32+arm64+aarch64+...
45.
by sfink 2mo ago
That's the missing part in the article. The example of the TSP solution for all cities in Germany was a pretty poor one compared to the optimal, which immediately raises the question: how good is it if you do some simple tricks to impr
46.
by sfink 2mo ago
Yes, that's understood, but it's not the point. That would convert a 3D space to a 2D space (the Z dimension is unaffected). But a 2D Hilbert curve converts 2D->1D, and a 3D Hilbert curve converts 3D->1D -- it's a curve
47.
by sfink 2mo ago
Why O(n log n)? I guess I thought that mapping a coordinate to its distance from the start of a space filling curve would be a constant time operation, or rather, it would only be proportional to the number of bits in your coordinate system
48.
by sfink 2mo ago
Anthropic is a US company, so it is lobbying the US to enact regulations to serve both its interests and the interest of safety. What else should it be doing? I imagine Anthropic (or at least many of the people who work there) are fully awa
49.
by sfink 2mo ago
I'd guess that's it, some kind of enforceable KYC. They'd only serve tokens to entities with a legally liable ID (as in, someone to sue, that would cost the defendant something nontrivial beyond a burner account or whatever.)
50.
by sfink 2mo ago
Wait, you don't do that already? I don't overcomplicate it, I just run ai-grep -v "bad code" on all of my source files and keep what's left. Why would you keep bad code around? If it breaks when I do that,
51.
by sfink 2mo ago
I'd like more people to be writing about this, because I find it fascinating. I have my own private project that is heavily LLM-coded, just to learn about what it's like. It's amazing how everything is different yet everythin
52.
by sfink 2mo ago
> What it actually proved is up for debate. I guess I'm debating, but it seemed clear enough to me? It proved that AI models and their harnesses are to the point where you can give them some work to do and leave them unattended for
53.
by sfink 2mo ago
Great, so any minor bug in Chromium's implementation of a spec is going to be forever embedded in the spec as the only allowed way to do it? Multiple independent implementations are the only practical way to ensure a self-consistent an
54.
by sfink 2mo ago
A lot of video-conferencing websites work that way because Chrome shipped a pre-standardization form ("Plan B") of WebRTC, it got a lot of uptake, and then they refused to unship it when the final standard came out with an alterna
55.
by sfink 2mo ago
The way I'm thinking of it is that they ported Bun from Zig to Unsafe Rust, and are still working on now porting it from Unsafe Rust to Rust (not 100% safe, but a reasonable amount).
56.
by sfink 2mo ago
But in my experience, that's what problem solving is like? You have a goal that you don't know how to get to. You come up with any way you can think of to reach that goal, and try out the ones you think might work. The effectivene
57.
by sfink 2mo ago
Ok, fair. I know that was also in the article, but your comment made me reread it, and I think you're right about the focus. I over-fixated on the headline. On second read, it's more like a lazy way to justify a preconceived idea
58.
by sfink 2mo ago
The article, summarized: there are incentives for OpenAI to claim that their AI hacked its way out of their network and into Hugging Face. Ok... and what? Do you have any evidence for the claim being either true or false that you'd lik
59.
by sfink 2mo ago
I agree with the attacker side. The actions of autonomous agents are absolutely the responsibility of the one or more humans that enabled them to take that action. Whether that means someone is arrested, maybe or maybe not, but at least the
60.
by sfink 2mo ago
At least Opus up to 4.7 or so, my experience is that Claude often uses "yes, with caveats" in place of "no, you numbskull". "Is a meter the same as a foot?" "Yes, exactly, you have it now, except they&#x27

More ›