fuglede_

Presumably, given that you also operate in the EU, you would have asked for their explicit consent before you did, so you could just check for that?

FWIW, Scott Aaronson has likened the question on factorization of concrete instances to what happened during the Manhattan Project:

> Joe #11: Then you’ll be waiting a long time, because that’s a stupid metric. (Or rather: anyone willing to use that metric today, is probably someone you shouldn’t trust.) It’s almost exactly analogous to judging the Manhattan Project in 1942, 1943, 1944 by the metric “how big an explosion can you make, today?” I.e., it’s something that’s going to be basically flat for a long time and then undergo a step change, under the assumption that everything is progressing as it should—because you need a critical mass in the one case and to exceed the fault-tolerance threshold in the other.

https://scottaaronson.blog/?p=4121#comment-1804101

And since all they need is a particular folder structure, you don't really need anything Git at all. Any kind of blob would do, cf. e.g. how Zenodo works.

It only works for GitHub; see https://palomar-registry.org/how-to-submit

And what a strange decision indeed.

One way to watermark (assuming temperature is otherwise positive) would be to output the most likely (or optimal) token every so often.

There seems to be something strange going on with how it plays with the GHCP harness: I've experimented on a variety of inputs (code, plain text, literature search), and more than half the time it falls into an infinite text/tool call loop a la GPT-2. Which is a bit spooky as you're still being billed for those infinite loops! But when it works it works, and responses do seem a good deal cheaper than OAI/Anthropic equivalents, so hopefully they'll get it ironed out.

But if you are going to self-host, then you can also self-host a Mastodon server and never have to migrate anywhere; wouldn't that achieve the same effect?

> Now I am migrating away

Where to, if I may ask? Are there any stable FOSS-emphasizing (preferably EU-based) alternatives (outside of just self-hosting)?

The new terms [0] say “You must not share projects that mostly consist of code written by "generative AI"-tools […]. Such projects having an unclear copyright status […]”

So okay, if you're concerned about the risk of liability from unclear copyright status, I suppose that's fine, but then why the “mostly”? It's not like the copyright status becomes less unclear by you supplementing with 50% hand-written code.

[0] https://codeberg.org/Codeberg/org/commit/96fac426a32d1ba91ff...

So why is this not the largest GDPR case in all of history?

Right; see Lakatos. In its roughest form, you study the structure of whatever counterexamples you find, add those as (negated) preconditions to your proposition, rinse and repeat until you have a true statement. If the proposition remains useful, you now have a new definition.

In fact, you don't even need an LLM tournament when you can have tom7's Elo World tournament: https://www.youtube.com/watch?v=DpXy041BIlA

3b1b just started a series under this slogan. Only one episode so far but chances are it'll be a good one; https://www.3blue1brown.com/

There are ways to make it work, but it requires using it only supplementary and with strict discipline. And it will look different for a 6-year old than a 13-year old.

But think of this trick we use about turning a tricky research paper in maths/science into something more tangible by making an LLM whip up an interactive version. That works at every level of education, and it means that you can completely tailor a piece of educational material to the kid

Tiny example: one kid was introduced to fractions and found it abstract that it was both about partitioning stuff and about numbers on a number line. So while we were practicing, I had an LLM make https://fuglede.codeberg.page/broeklegeplads/ to make it more hands-on.

Obviously for the small kids, this has to be an experience guided by teachers and parents, but for bright older kids with sufficient discipline that ought to be a useful trick for enhancing education.

Of course when we were kids, we would just write such educational programs ourselves and get the same effect /and/ learn to program (before getting banned from the computer room for putting spooky /binaries/ on the computers anyway), so maybe that's better for older kids. And maybe these kids will never have to do any maths or programming because the AI overlords have taken over when they grow up.

> I wouldn't be surprised if this was never acknowledged by Meta.

It will, at least, have to be acknowledged by making GDPR Art. 33 notifications.

Earlier: https://news.ycombinator.com/item?id=48324097

I've reached for this solution many times too; it's certainly convenient.

What I don't like is that all the resulting pages look the same. It's so easy to tell when a page is authored by Claude, and by now, I get the same “ugh another one” feeling as Dall-E generated images would give when they were completely overdone.

Maybe it's just be, but if not, maybe HTML generated outputs will also end up being dialed down from where they're at now.

But maybe not, since it's obviously a useful thing to be able to do. I wonder if there's a way out. To be able to introduce some natural entropy so not everything ends up looking the same. I guess not, since we're using machines whose natural mode of behavior is regression toward the mean. And maybe having it be harder to tell the author wouldn't even be desirable anyway.

So they're changing the product that people already paid an annual subscription for to the worse. That's asking for legal complaints.

It would be very helpful to know in understanding the capabilities of the models; and in getting intuition about where they are best applicable.

If the reason it was able to output the proof is that it happened to be included in an in-house university report written in Georgian, then that would make it less useful for research than if it's new entirely.

The XOR swap trick also features in the compilation/synthesis of quantum algorithms, where the XOR instruction (in the form of a CNOT gate) is fundamental in many architectures, and where native swapping need not be available.

One extension that I ran into, and which I think forms a nice problem is the following:

Just like the XOR swap trick can be used to swap to variables (and let's just say that they're bools), it can be extended to implement any permutation of the variables: suppose that the permutation is written as a composition of n transpositions (i.e., swaps of pairs), and that is the minimal number of transpositions that let's you do that. Each transposition can be implemented by 3 XORs, by the XOR swap trick for pairs, and so the full permutation can be implemented by 3n XORs. Now here's the question: Is it possible to come up with a way of doing it with less than 3n, or can we find a permutation that has a shortcut through XOR-land (not allowing any other kinds of instructions)? In other words, is XOR-swapping XOR-optimal?

I'm not going to spoil it, but only last year a paper was published in the quantum information literature that contains an answer [0]. I ended up making a little game where you get to play around with XOR-optimizing not only permutations, but general linear reversible circuits. [1]

[0] https://link.springer.com/article/10.1007/s11128-025-04831-5

[1] https://swapple.fuglede.dk

I've always loved this recording of Thurston talking about branched coverings and knot complements using big knots: https://www.youtube.com/watch?v=IKSrBt2kFD4