robryk @robryk@qoto.org

@landley @freemin7

TBH you can do the subterfuge with misskey-style quote toots too: just send a toot that refers to a nonexistent toot. Nothing prevents your instance from sending it out.

@danluu

Isn't there a large difference of degree? When I played around with chatgpt by asking it questions about mathematics, it usually would get the answer to the question wrong; not some tangential part of the explanation or an example, but literally the answer would be wrong. (When asked questions about physics or biology, it would be similarly badly wrong ~half the time.)

@danluu

This seems like a very effective condemnation of short essays as something that is intended to be graded, no? (We can see something that can respond to prompts for essays, but is bad at answering simple questions from the same domains.)

@wendypalmer

Similarly, there's the concept (that's apparent mostly in its falsification) that noun phrases actually refer to objects that can be described by each part of the phrase.

@ThatAdamGuy @Polychrome

In quite some cases companies could do the same thing without facial recognition: if they sell tickets to named people and verify official IDs. It's just that (good enough) facial recognition makes this available in more cases.

(BTW. There's a likely fundamental limitation of facial recognition that might come into play here, in that there's a lower bound on false positives, because the space of faces is not that large. This limits the number of banned people before the FP rate climbs too high to be acceptable, but I'm not sure if the limit impacts anything in practice.)

@techlife @lauren

If you can live with having to store a secret s.t. (a) anyone who owns it can impersonate you (b) you can't act as that identity if you lose it, then the thing you're asking for exists in secure scuttlebutt.

@techlife @lauren

> The network just needs to recognise changes in domain.

Try to figure out how that could happen (who trusts whom to make what statement). I don't think you get any stronger portability that what you have right now with old instance saying "this fellow is now foo@bar".

BTW. I've been only talking about authentication so far, but routing is an issue that also needs to be solved. I think that (as long as you actually want not terribly weak integrity guarantees) routing is simpler, and we have a couple of ways to do that in a distributed fashion (-> DHT, the most well-known example thereof being the thing that backs magnet links in BitTorrrent).

@dpiponi

So, we are starting to use ML models to mediate communication between people~ That's an uncommon start for "AI takeover" fiction :)

(this is not any real-world prediction)

@futurebird

Communities often lose their appeal when they become larger. This might be a (rather poor) reaction about attempting to limit growth.

Let's assume growth is inevitable in this case. The question then is what kinds of incentives one can create around joining to retain/gain properties of the community one cares about. Having joining require inventiveness surely promotes inventiveness in new joinees (and usually adds other requirements that are orthogonal to that, like amount of free time). However, technical complications _in the area of software_ do not really promote inventiveness or curiosity nearly at all past some critical point, because they can always be surmounted by following a guide. So, that creates a bias in the direction of more free time and some lower bar on patience -- neither of the two seem to me to be related to properties of the community one might wish to retain.

I would like to retain e.g. anti-outrage and gain more curiosity in this whole community. I don't expect technical hurdles to actually act in that direction, _even if they used to when the network was smaller_. Sadly, I don't know of any ways to actually promote them other than by example~

@TindrasGrove

The patent referenced in that article describes a machine that has a piston pump to spray water and has an open holding tank for water that's supposed to be heated from underneath. I expect the worst that can happen (modulo someone setting stuff on fire with the external heater) is very hot water being sprayed in the wrong direction.

Do you know if the pressure used by dishwashers increased early on? I frankly realized only now that I have no clue whatsoever even about the pressure my dishwasher uses.

@techlife @lauren

A quick skim of what they're suggesting shows that it's a setup where identity is longer lived than handles attached to it. Identity is of the "please hold onto this private key" kind. Handle-to-identity mapping is provided by otherwise authenticated organisations, who can revoke and reassign handles at will. So, organisation from whose namespace a handle is can repoint it however it pleases.

One interesting thing is that they seem to ask for some cert-transparency-style logs of handle-to-identity assignments, so that changes of those are necessarily publicly visible. (So, the expectation changes from "handles cannot be stolen/impersonated" to "stealing/impersonation of handles is auditable".)

You might be interested in how keybase's identity model works. It's remarkably similar to this one.

@techlife @lauren

Well, if we literally substitute it then "container for a user account" makes no sense (if the account is an _identifier_ then I presume it's public). This is why I think that explicitly speaking in lower-level terms can be helpful here.

@techlife @lauren

But then who has the authority to state what's a given handle's e.g. public key? Can they change their mind? (What if they disappear off the face of the Internet?)

@techlife @lauren

Note that the "your handle is a public key" setup exists, is well, and is used in some places. If handles are to be choosable and global:

Not everything is possible. Regardless of how many people are working on it, I will strongly doubt that e.g. they can accelerate something to more than speed of light.

I don't want to say that it's surely an unsolvable problem, because it's too poorly defined to assert anything like that. However, I doubt that a formalization of that problem that's close enough to be still subjectively considered the same problem is solvable. I expect that we'll use something that either doesn't have global chooseable handles or that adds some very strong assumption.

@techlife @lauren

Can you try expressing what you are saying using lower-level primitives? For example, can you rephrase this without talking about a "user account"?

Failing to follow my own advice, an interesting difference is that the "coin" "exists" beforehand and just changes the "owner", whereas a "user account" would be created out of thin air.

@Natanael_L @filippo @juliank

I don't get how verifiable delays are useful here. You obviously don't want to wait/burn compute to retrieve your secrets, so who would be burning the compute and when?

We define the rank of a matrix as the minimal number of rank-1 (i.e. of the form w^\dagger{}v) matrices that sum to it. I wonder what happens when we decide to optimize for something else: say, norm of type $foo over norms of type $bar over those matrices. Obvious candidates for $foo are L_{something}, for $bar are operator norms or Frobenius norm.

In particular, $foo=L_n and $bar=operator norm inherited from L_n seems potentially interesting (in particular for n=2).

#linearalgebra