My conclusion about that thought experiment is that it is trivially obvious that the system of books+human agent speaks Chinese even if the components do not. That's tautologically true. I think whether it understands Chinese is unanswerable at the moment. With our current level of understanding of intelligence, it quickly devolves into metaphysics.

Like you said, no individual neuron or even collection of 100K neurons in my brain can be said to really comprehend English, and yet the emergent entity that is me somehow does.

I think the only real conclusion is that until we really understand what intelligence is, to a better level than "I know it when I see it", then we don't really have any business dissembling about what entities are intelligent. Turing said as much when he defined his imitation game.

Because he's 10 and he wanted to try it out and it has a nice GUI for a lot of the tasks that Godot doesn't. He also wants to check out the thing that he sees in the splash screen for the games he plays. I don't want to discourage his interest by redirecting to something he might not understand and can't find as many video tutorials on.

He's learning C# in parallel, but in the meantime, until he learns to code, you can write an entire game loop in Unreal without writing any code at all, and that means he can see something and get feedback without wrestling with a compiler.

My son is trying to learn video game development, but you have to be 18 to download unreal engine, apparently, with no option to have a parent approve for you, d of course I helped him bypass it.

$10-$20 is still bargain basement rates to create a novel exploit.You only have to find it once to attack thousands of targets. For 1-off tasks like this where the baseline cost is so low and the payoffs so high, prices almost doesn't matter, within reason. It's daily coding use that a 10x price hike might crush.

But the online reseller can be fined for selling an illegal product in your state.

Around here it is rare enough for cops to ticket people blocking intersections, blowing through red lights, or doing 40 through a 20mph residential neighborhood. If CA is similar I doubt enforcement will be a priority.

The people who have shell access to the k8s hypervisor or worker nodes probably have root access anyway. Getting root access to the container pods does not buy you much beyond what you get by getting access to the service's shell account. At that point you can already read whatever secrets were injected via environment variables, read application log files, etc.

The main vulnerability strikes me as being on shared untrusted jump boxes, or like you mentioned, old school cpanel multiuser shell-based hosting environments where multiple untrusted users all have ssh access.

Common core and no child left behind has absolutely destroyed the education system. Because funding is tied to it, my kids spend WEEKS being drilled on EXACTLY what is on the test, not one iota more, with no thought to intellectual curiosity, no ability for the teacher to explore a tangent that gets the kids excited about the subject. Then the kids spend days grinding out meaningless standardized tests that don't go towards their grade when they could be learning.

And because the testing and the lead up to it is such a miserable shit show, there are usually a couple of "fun" days after or an early release to let the kids blow off steam, where again, no education is taking place.

How do you teach a literature class in a world of no homework? All class time would just be taken up reading the novel. Also forget about learning how to write an essay. We should just skip that too I guess.

Same with mathematics. You can learn the theory of how to apply integrals or trig formulae in class, but until you've worked a few hundred problems, you don't really know calculus or trig.

Same for foreign languages. You can practice speaking in class and learning about grammar rules, but you need to memorize how to conjugate verbs, exceptions, and learn vocabulary on your own time through rote memorization.

If you eliminate homework entirely class devolves to the teacher proctoring silent rooms of kids just being heads down doing the work they should be doing on their own time.

Class time should be spent on interactive activities with the teacher like discussing themes in a novel, teaching theory, hablando español, etc. Homework is for practicing and developing the skills you learn in class. Then you go back to class and receive feedback on the homework, correct misunderstandings, and move on to the next skill that builds on the previous lesson.

Does the average farmer need to own one? I imagine it must sit idle a lot of the time unless constantly in use. Could a regional services company own a few and process fields on a pay per use basis and spread the capital cost around?

Stunt: yes. Meaningless: no - there are now demonstrably fewer vulnerabilities in some software. Vulnerabilities that would have been discovered by malicious actors using the same process as Anthropic and exploited quietly had they not been publicly disclosed.

That's absolutely not true. Chattel slavery is by no means the only form of slavery. If you were to go back in time to ancient Rome or Greece or any number of other cultures, you would be surrounded by people who were slaves but were by no means born into it and may expect to become free again sometime in their life.

The US absolutely practices slavery in its prison system.

Sure they can, just stop buying it. The utility just disconnects their inverters and the panels become shiny glass. A solar cell without a load doesn't explode or anything.

We need more smart appliances that can be set to run on a signal from the grid. E.g. if I could plug my car in, but only have the EVSE deliver power when the grid tells it to. Or press a "delayed start" button on my dishwasher that tells it to run in 4 hours or when the grid tells it to, whichever comes first. Same for clothes washer/dryers.

If more houses had whole-home batteries, they could charge during the day and discharge at night to do load shifting.

I'm not the biggest AI proponent, but a security flaw is a flaw no matter who found it or how obscure. If the LLM agent can come up with an exploit that is demonstrable, then it should get fixed. That is not a scam, that is a real improvement to the security of the software under test. Who cares if nobody found them before? They are found now and so they need to be fixed now.

Like it or not, these tools are out there, and they are in the hands of state actors who are also using them to find exploits. If Anthropic wants to burn some of their money on finding and responsibly disclosing some exploits in software that is an important part of ur infrastructure, then great.