4 ms·
Computer Security - I'm firmly convinced that it has already been solved, but the solution[1] isn't widely known about or understood properly. Because we don't
by mikewarot 1y ago
Computer Security - I'm firmly convinced that it has already been solved, but the solution[1] isn't widely known about or understood properly. Because we don't have computers we can trust, we can't have communications that we can trust, and thus we can't have true Democracy.
Computing efficiency - The fact is that most of the transistors in a computer sit idle, waiting for their 15 nanoseconds of fame. A first principles review of what is actually possible leads me to suspect we can run LLMs and other large compute loads for 1% (or less) of the current energy requirements without new process nodes, using current fabs.
The Supply Chain - We need to have a publicly documented and tested second supply chain for everything in our society. Less efficient, and more robust ways to make every essential tool and material required for a modern society should be developed. Even if it's 1000x more expensive, it's still better to have that backup rather that being completely without options.
Radio Mesh networking - I'd set up a system in which everyone, everywhere, could stay in touch with everyone else. Governments and other organizations would handle the fiber backbone for big stuff, but a low bandwidth system that could work slowly, and globally, without it should be the thing we can all own.
[1] https://en.wikipedia.org/wiki/Capability-based_security https://en.wikipedia.org/wiki/Capability-based_security
- fsflover 1y agoIt seems your approach to security relies on correctness of the code, which is unrealistic in real life. In contrast, Qubes OS implements security through compartmentalization and seems to work much more reliably than any alternative.
- mikewarot 1y agoQubes is a capabilities based OS, for arbitrarily course grained capabilities. Far better fine grained systems are possible, there just don't seem to be any ready for prime time. Genode and GNU Hurd are about as close as you can get these days. Neither of them requires trusting application code, nor depending on its correctness. Think of it as using circuit breakers, instead of the current approach (in Linux, MacOS, Windows, etc) of just giving applications access to everything by default, and counting on them to be correct, and non-evil.
- fsflover 1y ago> Neither of them requires trusting application code, nor depending on its correctness. I'm not speaking about trusting the application code. This is obviously not safe. I'm talking about trusting as few lines of code as possible, including the OS itself. In other words, security-critical code should be minimized. For GNU/Linux, it's tens of millions of lines of code. You rely on their correctness. For Qubes it's about 100000: https://www.qubes-os.org/doc/security-critical-code https://www.qubes-os.org/doc/security-critical-code.