The paper misses a critical point. The A.I. may not self improve itself, but it may self improve based on the real world feedback loop.
Example: A.I. provider delivers bad models -> ai provider eventually gets less resources
Example 2: A.I. Provider delivers good models -> provider gets more usage and eventually more ressources
So, the system does not self improve but it gets the feedback from the rest of the world on how useful it is and also on how it can become more useful.
Useless. It rated ptv.de as pretty safe, where I built a replacement on, say 15 evenings using Claude code + OSM. The replacement is currently in early operations stage, but it nailed what ptv did before.
It rated the company where I work as completely replaceable. Which I rate as insane because if I think it could be replaced, I’d quit tomorrow and start a competitor that is cheaper/faster/better.
Someone will find a reflexive material to put on the drone. Then you have a multi kw laser that hits randomly anywhere when intercepting drones.
Also I wonder why it is not common to run interception drones that automatically fly towards incoming drones and captures them mid air. Like a wasp is capturing other insects.
So pretty much like the iron dome but not with single use rockets but reusable drones instead.
Shelly devices support json rpc over udp. It is quite nice to use those apis and the resulting client code is surprisingly light weight and robust.
However, there are two things that worry me:
a) what about datagrams bigger then MSS? Here the risk of losing the whole datagram doubles, triples, … Locally, it may not be so bad when one can use jumbo frames, but one the internet, it will be 1452 bytes per Datagram, before fragmentation happens
Their argumentation is pretty naive. If they would truely like to fix up the situation, they could simply put a tax on the exported electricity. Depending on the height of the tax, exports would be adjusted accordingly. Maybe, a non linear taxation scheme would be needed to allow exports in normal operation states and prevent exports in high price situations.
Example: A.I. provider delivers bad models -> ai provider eventually gets less resources
Example 2: A.I. Provider delivers good models -> provider gets more usage and eventually more ressources
So, the system does not self improve but it gets the feedback from the rest of the world on how useful it is and also on how it can become more useful.