Yes. The reward system for training humans seems to be similar to the reward system for training AI's in the sense that many (most?) people value an ill informed, confidently delivered opinion, if it confirms their biases, over accurate nuanced, qualified information, whether it comes from a human or a machine!

This.

Not really. If one partly unilaterally breaks agreements, there is no sense in negotiating new agreements. Some level of trust is required.

I think American's are deeply indoctrinated with the idea of American exceptionalism. So when others catch up or surpass America, it must be because everyone else is cheating or whatever. MAGA really taps into that, but it is not just MAGA. Can you imagine any US president being elected asking the people to accept that the American century is over?

Yeah, but if you have solar panels powering conventional A/C, already "the sunnier the day the more they work". It is not clear that some alternative direct heating approach would have better efficiency than solar cells plus electric pump based A/C.

I have solar panels and I rarely import power on the hottest days. What is more (for those without solar panels) the wholesale energy price is often negative during the hottest part of summer days.

It depends on what the "legal entity" is, but probably yes.
There are other ways to deal with that though. The land owner can require a surety or bond.

It seems that most human "reasoning" is rationalization after the fact. Humans are not very good at formal reasoning. Yes, we can be trained to do it, but most never receive that training and even those that do make most of their every day decisions in a "quick and dirty" way. The quick and dirty way is "sub-conscious" (though I think that term is going out of favor). Try 'Thinking Fast and Slow' by Kahneman. It is a "popular science" book, but the author is an authority. In any case we don't actually have conscious awareness of most of the processes that make up most of our reasoning. Most of what we pass off as reasoning is making up a plausible story for why the decision we made is correct. I see no reason why a LLM can't do that.

On the other hand, it turns out computers are very good at formal logic and have been for a long time. It is where early waves of AI enthusiasm focused. Look up "theorem prover". What they have not been good at is the the fuzzy heuristics of human intelligence.

An interesting question is whether you can gain the useful characteristics of human intelligence and retain the correctness of formal logic. I don't know, but it seems that the limitations of the formal logic approach are rooted in the practical impossibility of formally describing the world (ontology). See for example Cyc https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2F.... Classically, anything follows from a contradiction so if there is a contradiction anywhere in your ontology, then your formal reasoning is at best unreliable. Perhaps you can partition your system, so your neural network can talk to various tools: a calculator, a theorem prover etc, much as human can use these tools.

There's your mistake. Should have used emacs.

I can't make money selling at a negative price (unless I bought at an even more negative price), but I would want to minimize my losses.

For example, suppose I have a contract to buy a certain amount of power. The generator has built and managed a system to supply that amount of power, based, in part, on the contract they have with me, so I am bound to take that power. Now I can't on sell the power, because my customers have too little net demand, so I need to get rid of it. I expect the details will vary with the market, but think futures contracts.

Presumably provisions for decommissioning, both quantity and mechanism, vary from country to country. The question is, are they adequate? Costs 50, 80 or 100 years in the future are hard to estimate, even with good intentions, and proponents of nuclear power have an incentive to minimize them and have little risk that they will ever be held to account.

it does have ssh and everything you can do through luci you can do on the command line.

I'm not an expert by any stretch of the imagination, but my understanding is that standardized grammar is a relatively recent thing. After the printing press was invented, there needed to be a way to distinguish the "educated" from the "uneducated" and so lots of fiddly little rules were invented, many transposed from Latin. If you knew the rules your were "superior" to someone who didn't. Then they invented grammar schools to teach the rules to the right people.

Serious question: why doesn't this higher internal resistance cause the batteries to heat up as they discharge? I understand why cold batteries are a problem for stating ICE engines, but for an EV, wouldn't it just be a a short period of reduced efficiency until the batteries warm up? I have an EV but admittedly it doesn't get that cold here.

No all energy is not equal. Heat in particular is disordered energy and entropy matters. Heat can only be converted to other forms of energy, such as electrical or mechanical (work) by exploiting the difference in temperature between hot and cold reservoirs. Global warming results in very low grade heat. The maximum theoretical energy is low and generally impractical. You can do better if you capture and focus solar radiation, but generally solar PV is better.

This. LLMs are just an application. The core technology is deep learning.

Also, I find people confidently pontificating on whether AGI is achievable (ever or in some timescale) without having any real understanding of what these machines are, how they work or indeed what intelligence is.

The goal is: based on all past inputs, output an optimum, or near optimum response. The response might be a bit of text or it might be action of a robot or something else. A LLM might "just" be an advanced autocomplete function, but a sufficiently advanced autocomplete is intelligent (or indistinguishable from it) within the constraints of the modes available to it.

Think about it. An optimum response (by whatever criteria) requires a model of the universe. Training a neural network is causing it to build an implicit model of the "universe" - at least that part of the universe is has "seen" through its training data.