I guess we can add a whole new category to the Darwin Awards.

Isn't much of the point here the cultural shove? Sure, there's the line-go-up stuff; but that doesn't explain the companies gutting quite profitable software development operations to shovel money at Nvidia for things that have no demonstrated ROI; if it were nothing personal, just business, the level of enthusiasm for taking on poorly characterized risk would not be as fervent as it is. It's absolutely about resentment of the human resources that has been running at least as long as the demonstration that it would actually take some shoving to get them all to come back to the office, likely significantly longer.

It's the significantly less common case, for some combination of a lot of high-visibility stupidity being lethal and lethal accidents being public in ways that the details of nonlethal accidents aren't; but destruction of gonads under sufficiently foolish circumstances qualifies a living recipient.

They already pretty much are. You have to do at least a little performative fretting about the risks, which spoils the enjoyment of pure cheering at the best crunching sounds; but there's no way we'd justify the level of recreational head trauma something like football produces if we didn't fundamentally regard the players as relevant only the the way racehorses are.

AI video technology is still nowhere even remotely near just "click a button and take what it spits out". I don't know how to break this to anyone here, but you're not just going to go to some video generation site and turn out Woodnuts without extensive skill about AI video tools themselves and a wide range of traditional video production tools, and without spending weeks to months and significant financial expense on the project.

Even if / when this changes, video production is still always going to be limited by the human at hand. Most people's movie ideas, plotting, scripting, directing, etc frankly will be terrible. The slop in this case is the human, not the tool.

There are obviously cases where complete vertical integration makes no sense; literally all of them if you interpret 'complete' at full strictness; but when someone actually says "privatization" they basically always mean contracting out something large enough to be or have been an internal program. Sort of the way you don't say "outsourcing" unless it either was or plausibly could be an internal function. Ordering copy paper from staples or having a meeting catered generally doesn't count.

That doesn't mean to say that it's always a bad idea; but when someone says 'privatization' that's a "we'll have SAIC do it" proposal not a "employees and the DoE use laptops they got under a GSA schedule contract rather than from the First People's Computational Manufactury" proposal.

On the upside, AI lets anyone make a movie.
On the downside, AI lets anyone make a movie.

Including people who have terrible taste in plot, style, and everything else.

There's some genuinely good stuff out there - Gossip Goblin's work for example. But this is....

I'll just say, there's far better things that one could have spent half a million dollars on...

How are you defining "statistical inferences" as distinct from "logical inferences"? If you're defining fuzzy logic (e.g. not necessarily yes-or-no answers but allowing for ambiguity in conclusions), then we can agree conceptually, but your choice of wording is, I have to say, weird if so.

My understanding is that LLMs are built on a foundation of ANNs, and that indeed the backpropagation used to train ANNs is a statistical process;

Two responses. One, that's discussing individual-neuron scale processes rather than collective processes; and this was a discussion about inference, not training. Human neurons also learn by error minimization (Hebbian learning). But this does not describe the macroscopic processes that result from said minimization.

* During training, neurons develop into classifiers that detect superpositions of concepts that collectively follow the same activation process. Individual neurons weight their input space and subdivide it by a fuzzy hyperplane to achieve a classification result.

* In subsequent layers, said input space is formed from a weighted combination of the previous layer's classification; thus, the superpositions of questions being formed are more complex, as are the classification results.

* In a LLM, this iterates for dozens of layers, gaining complexity at each layer, to form each FFN

* The initial input space to a FFN is a latent (conceptual representation), as is the output; the FFNs, in result, function as classifier-generators; they detect combinations of concepts in the input space, and output the causally-resultant concepts into the output space

* FFNs alternate with attention layers dozens to hundreds of times in order to process the information, each layer building on the results of the previous one.

The word to describe that is not "statistics". It's "logic".

In a LLM, the first few layers focus on disambiguation. If there's a token for "bank", is this about a riverbank, a financial bank, banking a plane, etc? As the layers progress, it starts building up first simple circuits, and then progressively more complex circuits - you might get a circuit that detects "talking like MAGA", or "off-by-one programming errors", or whatnot. In the late layers, you have the general conclusions reached - for example, if it were "The capitol of the state that contains America's fourth-largest metro area is...", you've already had FFNs detect the concepts of fourth-largest metro area and encoded Dallas-Forth Worth, and then later taken that and encoded "Texas", and then finally encoding "Austin". And then in the final couple layers you converge back toward linguistic space.

Anthropic has done some great work on this with attribution graph probes and the like; you can detect what circuits are firing, and on what things those circuits fire, and ramp them up or down to see how it modifies the output. They very much work through long chains of logical inferences.

I use every style imaginable, including photos, in my tests. Same result every time.

One time I even did it with a Calvin and Hobbes comic, pretending than an AI made it. Responses included things like "The illustration also looks like shit and barely makes sense. Hope that helps.", "God damn this sucks so bad", "This also fucking sucks", and "The only punchline here is casual, pointless cruelty. if you think this is funny then you're literally a psychopath."

LLMs are not "statistical models" (randomness only even comes into play in the final conversion from latent space to token space because latent space is high dimensional, token space is low dimension, you need a rounding mechanism, and a "noisy" rounding mechanism works best; what you're thinking of, by contrast, is Markov models). And you cannot just "get lucky and randomly solve an unsolved math problem"; that's not how any of this works.

Also, it's silly that people are acting like "all problems but this one were already in the literature". AI has solved a whole slew on Erdos problems, and only a fraction had anything to do with existing literature.

And even in "existing literature" examples, it's not "nobody ever thought to search before" as if all mathematicians are morons, or that mathematicians adore putting out Erdos problem solutions without claiming them, It's that nobody had ever thought to apply an obscure technique from a given piece of literature to said Erdos problem.

The simple fact is, AI has gotten much better at solving unsolved math problems than humans are. It's simply another field that it's taking over, the same way it has been taking over programming. One can debate how much is "clever insight" vs. "just chugging away at possibilities until it hits on ways to advance toward the goal", but ultimately, that's a distraction from the fact that: it's getting really good at solving math problems that humans have spent decades on without success.

If they think it's worth it, on the balance, yeah. I'm not telling people where to put their money; or how to weigh the risk of Musk skimming the till on the actually-profitable rocketry in order to cover losses building mechahitler or trying to make orbital datacenters work.

I was mostly responding to the "I ask because I'd imagine the first thing the collective shareholders will ask for is that the crappy bits of Musk Inc. get divested as quickly as possible." part of the above poster's post. This is 100% an IPO where, to the degree legally possible and potentially beyond, and more so than even typical tech IPO voting structures, anyone who thinks that they are getting even a whisper of oversight, even at institutional investor scale, is kidding themselves.

It's up to you whether or not you think that a security representing whatever basket of endeavors Musk feels like conducting will be worth it; or if the risk that he'll bleed the winners to prop up his more dubious bets is too high; my note is purely that this IPO is genuinely rather novel in the degree to which it's designed to put the current CEO in effectively total control. Even compared to something like the mostly-unremovable Zuckerberg voting/nonvoting shares arrangement this has additional curbs on shareholder resolutions and litigation options.

Anyone who is hoping to avoid Musk's dumber ideas should probably just stay the hell away. Even by the standards of the classic "oh, there's actual-votes stock and peon stock; guess which kind we sold during the IPO" stuff; spacex is pushing things. The boy-king of mars holds 85% of the voting power; shareholders are required to waive the right to jury trial or class action and submit to arbitration only, only class B shareholders(mostly Musk) can remove the chairman of the board, CEO, or CTO; and similar enthusiastic use of Texas' provisions for 'controlled companies' that really don't want to take any pesky outside input.

Putting your money on that is pretty much entirely just making a bet on whether you think the dictator for life will make line go up or not; not even pretending to be analogous to an ownership stake.

If it weren't a technical issue; would it actually be a moral issue?

I suspect that there are lots of ways, including some surprises, of getting the problem wrong to some degree and introducing nasty developmental issues; so I could see an IRB having very plausible objections to the "eh, we'll keep pumping out flipper babies until we trial and error our way to what an embryo requires to develop a brain stem properly!" R but, if for sake of argument, you had a system that actually worked wouldn't that basically just be surrogacy without the seedy undercurrent of economic conscription?