In the end, it's still a somewhat better search than Google (If it gives you links to original sources, which a lot of the LLMs do).

Imagine having such an immense vocabulary, and using it that weakly.

The capacity of the government of a large jurisdiction like California, or more particularly the US, could bankrupt someone like Musk, so I say, bring it on. Within a decade Musk would have abandoned all efforts, or, even better, be stone cold broke (frankly billionaires shouldn't exist at all, and we should tax the living fuck out of them down to their last $200 million).

We're too afraid of these modern day Bond villains when we should be aiming every financial, and probably every real, cannon straight at them and putting them in a sense of mortal danger every minute of their waking lives, so that they literally piss themselves in terror at the though that "we the people" might decide to wipe them out for good.

Even if the prediction of comparatively controlled impact is accurate; I think it's worth considering just how grim it is likely to be; not in purely economic terms; but in the character of the work.

Maybe this is a personal peculiarity; but I that there's something exquisitely dispiriting about beating your head against people who are stubborn or clueless enough that every conversation is just a baffling sequence of different confusions, some of the repeated from previously. It's a totally different thing from dealing with someone who is merely ignorant; but learning, especially if they are enthusiastic about it.

Even if everything is fine in terms of job pace and security and all; that seems like it is shaping up to be a really hellish aspect of dealing with bots. The experience is sort of a somewhat weirder simulation of dealing with a chirpy, people-pleasing, very-junior type; except they are far more likely to lie than to admit ignorance; and they never learn(possibly the SaaS guys hoovering up your interactions in the background will make the next iteration better, possibly not, progress seems to have slowed considerably after only a brief period of improvement; but a given release is more or less full groundhog day).

That seems like a nightmare. Everything that sucks about teaching or mentoring; but precisely none of the rewarding aspects.

It's adorable how they pretend that the 'well being' gap between the people who matter and the ones who don't is some sort of surprise that calls for urgent action; rather than a deliberate outcome carefully achieved.

It's the pandemic-period numbers that are the anomaly, from a period when at times downright existential issues forced people's hands(at least for white collar workers; if you are 'essential' good luck and back to dealing with the public in person); and a lot of work has been put into rectifying that period.

What's next; a comparative analysis of the labor markets of the 1950s and the 1980s that studiously pretends that it's not exactly as Milton Friedman and Neutron Jack intended?

In a sense his puzzlement is justified; when the tech demo works an LLM is probably the most obvious candidate for 'just this side of sci-fi'; and, while may of the capabilities offered are actually somewhat hollow (realistically, most of the 'take these 3 bullet points and create a document that looks like I cared/take that document that looks like my colleague cared and give me 3 bullet points' are really just enticements to even more dysfunctional communication) some of them are fairly hard to see duplicating by conventional means.

However, I suspect that his perspective is fundamentally unhelpful in understanding the skepticism: when you are building stuff it's easy to get caught up in the cool novelty and lose sight of both the pain points(especially when you are deep C-Level; rather than the actual engineer fighting chatGPT's tendency to em-dash despite all attempts to control it); and overestimate how well your new-hotness stacks up against both existing alternatives and how forgiving people will or won't be about its deficiencies.

Something like Windows trying to 'conversational'/'agentic' OS settings, for instance, probably looks pretty cool if you are an optimism focused ML dude: "hey, it's not perfect but it's a natural language interface to adjusting settings that confuse users!"; but it looks like absolute garbage from an outside perspective both because it's badly unreliable; and humans tend not to respond well to clearly unreliable 'people'(if it can't even find dark mode; why waste my time with it?); and because it looks a lot like abdication of a technically simpler, less exciting, job in favor of chasing the new hotness.

"Settings are impenetrable to a nontechnical user" is a UI/UX problem(along with a certain amount of lower level 'maybe if bluetooth was less fucked people wouldn't care where the settings were because it would just work); so throwing an LLM at the problem is basically throwing up your hands and calling it unsolvable by your UI/UX people; which is the an abject concession of failure; not a mark of progress.

I think it may be that that he really isn't understanding: MS has spent years squandering the perception that they would at least try to provide an OS that allowed you to do your stuff; in favor of faffing with various attempts to be your cool app buddy and relentless upsell pal; so every further move in that direction is basically confirmation that no fucks are given about just trying to keep the fundamentals in good order rather than getting distracted by shiny things.

Lincoln was a Free Soiler. He may have had a moral aversion to slavery, but it was secondary to his economic concerns. He believed that slavery could continue in the South but should not be extended into the western territories, primarily because it limited economic opportunities for white laborers, who would otherwise have to compete with enslaved workers.

From an economic perspective, he was right. The Southern slave system enriched a small aristocratic elite—roughly 5% of whites—while offering poor whites very limited upward mobility.

The politics of the era were far more complicated than the simplified narrative of a uniformly radical abolitionist North confronting a uniformly pro-secession South. This oversimplification is largely an artifact of neo-Confederate historical revisionism. In reality, the North was deeply racist by modern standards, support for Southern secession was far from universal, and many secession conventions were marked by severe democratic irregularities, including voter intimidation.

The current coalescence of anti-science attitudes and neo-Confederate interpretations of the Civil War is not accidental. Both reflect a willingness to supplant scholarship with narratives that are more “correct” ideologically. This tendency is universal—everyone does it to some degree—but in these cases, it is profoundly anti-intellectual: inconvenient evidence is simply ignored or dismissed. As in the antebellum South, this lack of critical thought is being exploited to entrench an economic elite. It keeps people focused on fears over vaccinations or immigrant labor while policies serving elite interests are quietly enacted.

I suspect that a more fundamental problem is what you would need to preserve.

Embryos are clearly the easier case, being small and impressively good at using some sort of contextual cue system to elaborate an entire body plan from a little cell glob(including more or less graceful handling of cases like identical twins, where physical separation of the cell blob changes requirements dramatically and abruptly); but they are also the case that faces looser constraints. If an embryo manages to grow a brain that falls within expectations for humans it's mission successful. People may have preferences; but a fairly wide range of outcomes counts as normal. If you discard or damage too much the embryo simply won't work anymore; or you'll get ghastly malformations; but there are uncounted billions of hypothetical babies that would count as 'correct' results if you perturb the embryo just slightly.

If you are freezing an adult; you presumably want more. You want the rebuilt result to fall within the realm of being them. That appears to not require an exact copy(people have at least limited ability to handle cell death and replacement or knock a few synapses around without radical personality change most of the time; and a certain amount of forgetting is considered normal); but it is going to require some amount of fidelity that quite possibly wont' be available(depending on what killed them and how, and how quickly and successfully you froze them); and which cannot, in principle, be reconstructed if lost.

Essentially the (much harder because it's all fiddly biotech) equivalent of getting someone to go out and paint a landscape for you vs. getting someone to paint the picture that was damaged when your house burned down. The first task isn't trivial; but it's without theoretical issues and getting someone who can do it to do it is easy enough. The second isn't possible, full stop, in principle, even if you are building the thing atom by atom the information regarding what you want has been partially lost; though it is, potentially, something you could more or less convincingly/inoffensively fake; the way people do photoshop 'restoration' of damaged photos where the result is a lie; but a plausible one that looks better than the damage does.

The fraught ethics of neurally engineering someone until your client says that their personality, memories, and behavior 'seem right' is, of course, left as an exercise to the reader; along with the requisite neuropsychology.

It's different from humans in that human opinions, expertise and intelligence are rooted in their experience. Good or bad, and inconsistent as it is, it is far, far more stable than AI. If you've ever tried to work at a long running task with generative AI, the crash in performance as the context rots is very, very noticeable, and it's intrinsic to the technology. Work with a human long enough, and you will see the faults in his reasoning, sure, but it's just as good or bad as it was at the beginning.

Correct. This is why I don't like the term "hallucinate". AIs don't experience hallucinations, because they don't experience anything. The problem they have would more correctly be called, in psychology terms "confabulation" -- they patch up holes in their knowledge by making up plausible sounding facts.

I have experimented with AI assistance for certain tasks, and find that generative AI absolutely passes the Turing test for short sessions -- if anything it's too good; too fast; too well-informed. But the longer the session goes, the more the illusion of intelligence evaporates.

This is because under the hood, what AI is doing is a bunch of linear algebra. The "model" is a set of matrices, and the "context" is a set of vectors representing your session up to the current point, augmented during each prompt response by results from Internet searches. The problem is, the "context" takes up lots of expensive high performance video RAM, and every user only gets so much of that. When you run out of space for your context, the older stuff drops out of the context. This is why credibility drops the longer a session runs. You start with a nice empty context, and you bring in some internet search results and run them through the model and it all makes sense. When you start throwing out parts of the context, the context turns into inconsistent mush.

Is this the same Klarna that has already been through one cycle of being super bullish about AI and then having to re-hire some actual customer support?

It shouldnâ(TM)t really be surprising that the compulsive liar was lying.

That's only very partially true. The uptick in unpaid mortgages gave the house of cards a little tap; but it was the giant pile of increasingly exotic leverage constructed on top of the relatively boring retail debt that actually gave the situation enough punch to be systemically dangerous; along with the elaborate securitizing, slicing, and trading making it comparatively cumbersome for people to just renegotiate a mortgage headed toward delinquency and take a relatively controlled writedown; rather than just triggering a repossession that left them with a bunch of real estate they weren't well equipped to sell.

My workplace allows browsing outside of the Intranet with Chrome, Firefox, Safari, maybe others... basically anything but Edge.

That's not the question.

The question is whether the AI-produced code is a derivative of existing code, and the answer is still not resolved.

In some cases, the answer is a clear YES, because the code is a direct copy of something written by someone else. If something like that ends up in the kernel, it will have to be removed when someone notices.