skitsofrandom

I’m wondering if AI has made SIMD intrinsics much more approachable for many and so there are just more people working on abstractions for their workflow of choice right now. There’s probably a lot of code out there that could benefit from SIMD but the effort to actually use it was too high for the return.

There are a lot of documented small scale hacks on American infrastructure by state actors. The prevailing theory I have heard was that the point is to find as many vulnerabilities in our internet-connected critical infrastructure, then sit on it until it’s needed. It doesn’t make news when a few water tanks overflow in a random tiny Texas city, but the principle is that these kinds of exploits may generalize to the infra of major cities.

Example: https://texasstandard.org/stories/russian-hackers-muleshoe-t...

So there is no need for a highly visible active cyber war. We have tons of cyber professionals working on this stuff day in and day out to help keep up prepared.

Well but the point goes beyond “do they let you through.” All of this goes past the here and now. It’s about getting flagged in some system that causes you to get picked out 3 stops down the line when these odd occurrences add up to you not being trustworthy based on your profile history.

That kind of history is already being collected about people. That’s what you should be worried about when it comes to engineering some scheme that sounds clever.

This may feel like a good idea as a “gotcha” justification but it just doesn’t matter. It’s still extremely abnormal and you will stick out. The only way to protect yourself is by blending in, not sticking out.

I worked on a team that did this, + required feature branches to get squashed to a single commit before merge. On that commit, we'd require a very brief bullet list of changes, sometimes a POC, and a link to the PR. This made git blame a lot more helpful when debugging issues. I am a big fan of this approach.

Yeah but if you're a normal guy strolling through every time with a phone that has nothing- no pictures, no signed in email, no history of messages, 4 contacts. That's abnormal, no way of spinning it as "but I just don't use my phone much" will make that seem normal. The average person has their phone glued to their body 24/7 now. Implying that you don't is abnormal.

It's easy to get all high and mighty but there really doesn't have to be complex subterfuge behind "replacing the majority of our technical infrastructure with devices solely created within the borders of our primary military, industrial, and economic rival is not a good idea for security and sovereignty."

We cast aside local manufacturing for cheaper prices in another country and are going to pay the price one way or another.

I would guess a good way to implement a language meant for LLMs would be to strip out as many features as possible, so that all programs needed to be composed of just a handful of common structures written in shorthand. As much compiletime enforcement as possible. Maybe human-crafted abstractions for more error-prone concepts like multithreading that limit how they can be used in the language but provide ample, testable bumpers.

It would be interesting to design some mechanism where an external actor could program extensions into the language easily, which then would become a new black-box tool that an LLM could program with. Ideally though, I think you want to remove as much agency in reimplementation and re-abstraction as possible from the LLM. This is really just some tier above current high level languages, with as much dumbing down as possible.

I'm not sure this is really necessary or would even be effective because there wouldn't be a million examples for the training data of an LLM. But I imagine something along these lines could make an LLM more productive and token efficient.