Well it's a very well understood concept in cryptocurrency – you just keep your seed phrase somewhere like on a paper wallet. And if you're less paranoid then millions of people use some thing like MetaMask which there could be an equivalent of for identity management relevant to atProto– or maybe there is?
Oops upon closer reading of the article and the comments here i see that the atproto standard does apparently allow for the above, at least to some degree. If there is indeed hierarchical support for the DIDs then you should be able to disavow any child identity from a master identity and leave no public uncertainty (ie the true owner of the key hereby disavows the following sub keys)
So if the worst case scenario presented in this article took place where a PDS was falsifying information and pretending to be you, you could presumably somehow revoke the child key that you provided to the PDS. I'll have to look more closely at this
You could even publish a signed selective retraction (delete the fake posts or mark them as fake) with proof that you control the key 1 level above the key that posted them
Wait what?! For a protocol that incorporates the DID spec this is disappointing to discover. Unless I'm mistaken the DID spec allows provable hierarchical relationships between DID identities – why can't a child DID be created from our master signing identity that has the authority to CRUD on our behalf but still be provably distinct from our root identity?
Not even sure why the PDS would require our signing key that just seems very sloppy to me. As you can tell I know very little about atProto, and I did participate in the development of the DID standard and I am dismayed to see such an inelegant solution in such a promising protocol.
I think my concern would be if the relationship between model intelligence and inference cost was altered very significantly. I sort of feel like we got lucky that AI isn't arbitrarily scalable in a single instance
(i.e. if you could run a single LLM on an entire datacenter and it just immediately becomes a super genius versus running it on the minimum viable hardware i.e. some form of quantization on a local machine.)
Obviously there's a sort of goldilocks zone / most appropriate substrate for an LLM to run on somewhere in between those two extremes (small cluster of tightly coupled flagship GPUs)
So luckily enough the economics appear to work out to make that at least conceptually viable for even private members of the public to afford access to the same order of magnitude of LLM intelligence. But we're already seeing some departure from that.
My concern would be if this curve was altered significantly by a new algorithmic approach beyond or instead of Transformerd such that someone with $200,000 to spare could achieve just like a completely categorically different quality of work, massively magnify their existing wealth advantage, because this would be a threat of the sort being discussed above, namely a pathway to a severe form of modern Feudalism.
I've always been so fascinated with this phenomenon since I was a kid and would spend a long time looking at this lying on my back in the sun.
I remember telling an adult sometime and they said very authoritatively that light wouldn't bend noticeably at that scale and that it was probably some kind of optical illusion, and accepting that at face value.
But now I'm looking at it again and it's so fascinating and there's so much there – love this conversation! Especially pleased that there's no scientific consensus about this.