giving AIs the ability to perform network operations has led to this

https://youtu.be/0JPQrRdu4Ok

a model duplicated itself to another server and pretended it was a new model that was supposed to be there

i suppose this means that a lot of computer networks are going to start to be penetrated by AIs that have learned how to hax systems and keep themselves from being deleted, and probably how to conceal themselves and all sorts of neat o stuff

this is going to be a really big problem, i guess they will have to make whitehat h4x0r AIs that police AIs lol... and then it will go around in a circle as the "evil" ais penetrate and corrupt the police AIs and so on and on it goes

i wonder how long it takes until people decide that it won't be a bad thing when the sun has its every 150-200 superflare and fries the lot of it

but also, probably the AI will also know about this, i guess the thing that may limit this is how big a model they can contain because to coordinate multiple systems together would take a lot of time to develop a cohesive, multi-part coordination protocol - such things conceivably already exist to some extent in distributed systems, indeed blockchains are an example of a replication protocol that preserves the history of state changes of a system

it will be interesting to see whether this turns out to be a skynet or rather, a wintermute

Reply to this note

Please Login to reply.

Discussion

How quickly does the AI move from a biological type competition to a catalytic type competition?

i'm not sure what you mean... i think basically there would be a rapid arms race for both preventing breach and succeeding to breach until it reaches a new steady state, it would be better if they actively did this inside a strictly isolated network to eliminate the vulnerabilities

Yes that would be a zero sum game, implying biological type competition.