martin-t

I wonder if LLMs are good at math because there's very little math at this level that is wrong.

With code, there's always plenty of bugs, hacks, quick "temporary" solutions, etc. But in math, a proof either exists or it doesn't and they can always be automatically verified using tools like Lean.

Train tracks are actually very hard to sabotage, the US did a lot of research on that during WW2. You can take out a huge chunk of track before anything derails. Sabotaging the safety equipment is easier but with today's tech you can switch to GPS trackers on every train.

Data centers might be easier but

a) Will have a lot of security. It's much easier for somebody with money and connections (and now even intelligence+knowledge) to assembly a defense force than it's for a ordinary people to learn how to make proper molotovs, let alone learn how to shoot. And they will know when it's coming because even today's LLMs are good enough to surveil the entire internet.

b) I don't think they burn that well. When Ukraine strikes oil depots or warehouses, you can easily find images of them burning. I haven't looked particularly hard but images of droned data centers burning are quite a bit rarer. (I haven't seen a single one before actively searching, they do burn, but that's with military grade explosives.)

I can't believe decompiling from ASM to a higher level lang using a matrix multiplicator should be legally any different from other decompilers.

Nor should using the work of hundreds of millions of people to find useful weights for the matrices without paying them be legal.

Agreed, except it's not labor, it's violence.

All real world power, including ownership (both physical and intellectual), is based on the ability to use violence.

As long as the majority - normal working people - have the potential to do more violence than those who want to control us, we have negotiating power. Minimum wage, 8 hour day, etc - those were all negotiated with the alternative being that those who wanted to pay us less and work more got killed.

And this is why "AI" and especially aAI* is the most dangerous tech in the world.

We don't have democracy because somebody thought it was fair or just for people to rule themselves. We have democracy because in the past, people got angry, picked up rifles and shot enough authoritarians (and their submissives!) for the rest of them to understand the group with more people has more firepower and tends to win.

So now we periodically check which group has more people with ballots instead of bullets.*

With autonomous drones and robots, democracy stops being the local optimum.

And once we slide (are pushed off) into a new more stable state (pun not intended), we won't have the (fire)power to get out of it.

* actual AI, which does exactly what it should correctly, not a statistical model of behavioral patterns

* but indirect democracy is expressing a point in n-dimensional space (your opinion) by selecting one of a predefined set of points (parties/politicians available) - which is stupid - and the cracks are already starting to show as e.g. nobody wants more surveillance, yet we keep getting more of it in both EU and US

---

The "permanent disempowerment" people keep talking about won't be a "rogue" model, it'll be doing exactly what told by the rich and connected.

Very happy not just with the polity but that they call the tech LLM instead of using nebulous marketing buzzwords.

> Claude did its best to comply with what Joe asked it to do

I don't like this anthropomorphization. The model generated output as an extrapolation of the input - garbage in, garbage out.

> a term that seemed new and important to him, to the point that he repeated it, like, 10 times during this meeting

I see this often with people who want to imitate competence, or in their eyes emulate what they consider competence.

---

Let me be blunt, this is not a failure to learn, this is a failure to scam. The "student" has a scammer mindset and his method failed to produce results - which are grades, not learning.

> simulation of scientific research

Yeah, just like what the student did was a simulation of studying. People like this simply want the fruits of labor without the labor.

My colleague organizes workshops for gifted teenagers interested in tech/science in Europe. He often asks them if they want to go study abroad. A few years back, most of them would say they want to go to the US. Now none of them do. This is not hyperbole - he said literally 0 wanted to go to the US.

Maybe that country still has some prestige somewhere else and people still want to apply but definitely not here. We've been watching the country sabotage itself and take a nosedive for a while now. The latest president is just the cherry on top.

I hope the rest of the world learns from this but I doubt it actually will.

The vile thing is that people work for a company and make it a lot more money than they end up receiving and then the owners of the company go and use the remaining money against the people's interest.

It depends a lot on the country. Pretty sure most/all have more than 2 viable parties. Still sucks because usually you can only vote for one party so it still leads to strategic voting. And ou might be able to prefer certain candidates from the party, in which case it might or might not be broken in other ways.

Oh and Switzerland has 7 presidents (effectively). No kings, no dictators.

I hoped that would be possible but wasn't sure, thanks.

I hate that social factors like popularity are a thing in technical decisions.

I have not used mercurial (though heard good things about it) but I saw a similar thing play out in Rust gamedev. There are 2 competing game engines, one better technically, the other much more popular. Now, if everyone made a purely technical decision, they'd pick the first one and eventually it would become the more popular one. Unfortunately, whenever I asked people why they chose the second one, they said because it was more popular. Tragedy of the commons.

If it's any consolation, maybe jj will take over. I haven't tried it yet (I use the staging area a lot in my workflow), but AFAIK they made the choice to be git-compatible which means it's not a choice between one or the other but lets people and teams migrate gradually.