I have a student who is writing a paper about exactly this topic. Almost any large project nowadays uses dozens of external libraries, which in turn use dozens or hundreds more. This creates a huge, almost unknowable dependency tree. Any of those libraries may be updated at any time, and be pulled into a new release of your software. Any of those libraries may contain a security flaw that could be discovered and exploited. Any of those libraries may be deliberately compromised - and how would you know?

As a current example, consider the recently discovered flaw in Starlette, which the developer claims is downloaded 325 million times per week. Never heard of Starlette? That's because it is a fundamental building block buried deep in that dependency tree. Despite the title of the article, this flaw affects far more that just AI apps.

IMHO, the best solution - if you can afford it - is to write as much of your own code as you can. Sure, you may also have security flaws, but you are a far smaller and less interesting target. If there is a better solution, I don't know what it is...

The US government can compel any US company to release data that it holds, even if that data is stored outside the US. Pretending that any US company can comply with the GDPR is a fantasy.

This might, might be acceptable, if one could trust the US government. At latest after the Snowdon revelations, we all know that you cannot.

What is it with the addiction our governments have to mass surveillance? Were the police unable to do their jobs before the Internet? When correspondence was by physical mail instead of Email? When they had to get warrants before invading a person's privacy?

That's an excellent write-up by curl. It shows that the entire reporting system needs revised, in the age of AI reports. Automatically assigned the highest scores to vague reports can no longer be justified.

This seems slow. Current radio technology exceeds that by 1 to 2 orders of magnitude. Plus, lasers tend to have problems with cloud cover.

What am I missing?

Paying ransom ought to be criminal. Moreover, they have zero guarantee that the data is really gone, no copies anywhere.

The problem with this kind of trial is that it's all about personal motivation, and personal memories. Insight into motivations is difficult at the best of times, and there is little way to prove them.

Memory is worse. Human memory is fallible. Especially in cases of conflict, we unconsciously edit our memories to cast ourselves in the best light, and our adversaries in the worst light. As a personal example: We have a couple next to us who are a$$hole neighbors, who have (imho) deliberately sought conflict with us multiple times. At one point, i went back to the correspondence we had on one issue and...it was very different than what I had "remembered". They were still jerks, but my memories had morphed to make things far more black-and-white than they actually were.

So Musk saying what Altman wanted, and Altman saying what Musk wanted - you can believe as much of it as you want, but likely very little of it is accurate. Remember that there are three sides to every story: What person A remembers, what person B remembers, and what actually happened.

The industrial revolution saw a huge shift of workers from agriculture to factories. Transitions are always hard, and factory working conditions were not always the best. Still, over the course of a generation or two, the industrial revolution lead to a huge increase in the average standard of living.

AI has exactly this potential. We are still in the very early days, seeing some of the initial pains of transition. However, the potential of an equally huge shift is definitely there.

Ransomware would completely die out, if people simply refused to pay. No profit to be had, criminals would spend their time elsewhere.

As encouragement, paying ransom should simply be illegal, with severe personal penalties for any administrator or managers who approves such a payment.

Your comment about administrators is absolutely right.

I'm in Europe, where the problem is less pronounced. Still, over the last 20 years, the ratio of non-teaching staff to teaching staff has gone from 2:3 to 3:2. Those numbers don't look dramatic, but consider: It used to be that 100 teaching staff had 66 admin staff. Now that same 100 teaching staff have 150 admin staff, so 2.5 times as many. Not that our teaching loads have been reduced - much the contrary - our classes are now larger. You have to fund the bloat somehow.

I am reminded of the famous quote: "The bureacracy is expanding, to meet the expanding needs of the bureaucracy."

So most houses have fatter wired than they usually use. But comes the day they do use the capacity, and...what happens? Plus, you now have to maintain hardware distributed over zillions of different structures. Plus, you have to build those structures. I can't see how this can possibly be cheaper or more practical.

"it is common for liberals to do things like put up signs in their yards that say they stand with the homeless while simultaneously voting for zoning policies to defend their property value"

So much this. Where I live, the politians are pushing to house 200 illegal African and Middle Eastern immigrants in a town of around 1000 people. Let's be real: that will destroy the town.

Why not house them in the affluent suburbs where the politicians live? We all know that will never happen...

"Everyone's fine with security cameras"

Um, no. In many, many countries, including mis if Europe, you cannot have cameras aimed at public spaces. For exactly the reasons listed. People moving in public spaces have a right to do so without being tracked.

Machine faster than human. They may also be physically stronger. The only thing this shows, is that humanoid robots have become reasonably efficient (assuming no recharge breaks).

This may be fancier than captchas, but it is arguably even easier to fake. If I don't have someone's Iris scan, then a hacker can send anything that looks legit. If I do have your data (um, oops?) then a hacker just has to get ahold of it.