At least half the fun of being a 'disruptive innovator' is being able to treat your abject ignorance of history as a strength rather than a deficiency.

The two aren't actually so different. You do get to make economic arguments a lot more openly about copyright(while, when it comes to killing, we normally make them relatively quietly and circumspectly when the unpleasant matter of what risks to the public are just part of The March of Progress and which ones are negligent or reckless comes up. We prefer not to talk about it; and have some proxies like 'VSL/ICAF' to help; but we do it); but the classifications are ultimately a policy thing and open to amendment as needed.

"Murder" superficially resembles a stable category just because of a true-by-definition trick: we call it 'murder' if a killing is unlawful and forbidden(or, rhetorically, if we think it ought to be unlawful and forbidden); so there's always a strong anti-murder consensus because everyone is against killings that are forbidden, except a few Raskolnikov-type edgelords. What there is not is actually a consensus on what killings we are or aren't against. The people who think that every other defensive option must be exhausted and the ones who are just itching to castle-doctrine the next fool who steps over the property line are both anti-murder; but not entirely in agreement on what that means; same with the current dispute over whether euthanasia is a legitimate exercise of self determination or nihilistic hyper-sin; or any of the wartime arguments over where 'collateral damage' stops being unfortunate-but-proportionate and just goes into being bulk murder.

It is somewhat more common to find(in public, not so much remotely in the vicinity of legislative power) people who will outright claim to be against copyright; because they do not consider any derivative works to be legitimately unauthorized; but here it's a more or less straightforward fight between two entities that would both claim to be in favor of copyright; but who differ on whether setting up a data mine in the BBC's backyard is copyright infringement or not.

He's the founder and CEO of one of the companies HP bought during Apotheker's...impressive...string of failures. That was in 2011; but it remained in the news first when HP wrote down their 10.3 billion dollar buy by 8.8 billion dollars; then when the litigation began by HP against previous management on the theory that they must have been cooking the books a bit for things to go so wrong so fast under HP's illustrious management.

The charges stuck against the CFO; but the CEO and VP of finance were acquitted. Then the VP of finance got hit by a car; and the CEO's celebratory yacht outing took a literal turn when the ship capsized and he died; then the VP of finance finished succumbing to his head injuries and died less than 48 hours later.

I'm not sure anyone thinks well enough of HP's ability to execute to seriously suspect them; but the background probably didn't reduce interest in getting a nice decisive root cause for the boat issue.

Obviously the only computer experience worth having is on a Copilot+ AI PC; so microsoft is doing us a favor by helping us avoid legacy experiences.

Yeah, I'm hung up on that too. You can come up with some outrageously huge numbers for mass and angular velocity, but once I multiply them by zero distance... I'm missing something.

the outer edge of the mass exceeding the speed of light

That intuitively makes sense, but I thought part of the black hole cheat is that it doesn't have an edge. I thought they were literally singularities, with a circumference of zero. Apparently not the case?

How a thing with a circumference of zero could meaningfully "rotate" is beyond me, but I thought this (and many other suspected properties of rotating black holes) was supposed to be beyond my ignorant layman understanding!

What they are not saying, but what I suspect is part of the plan, is that the numbers only work out if you actively move toward being more locked in to their products.

If you just want the basic vsphere-managing-a-few-ESXi-hosts setup the bundle is stupidly expensive; but if you try to justify the cost of the bundle by using other parts of it you end up with a system that is significantly harder to migrate away when Broadcom decides to alter the deal further.

That's not how that kind of Thunderbolt device works:

TB handles the PCIe tunneling; but the PCIe device still acts more or less as an ordinary PCIe device would. On the plus side that means that TB can support basically arbitrary PCIe peripherals(barring some very fiddly assumptions most commonly found in GPUs, especially if the firmware or OS expects a GPU to be available very early in boot); but on the minus side it's still up to the OS to know what to do with the PCIe device.

USB was a slightly wierd one because it was very much an Intel pet project. Not sure if there was any arm twisting involved in terms of chipset pricing or similar; but they quite deliberately made the then-ubiquitous '2 USB 1 ports forlornly doing nothing' a cheap to implement default chipset feature; rather than USB being a 'premium' chipset upsell or a discrete controller(obviously those existed as well; but only really became relevant once people actually wanted USB ports).

As a result the standard for typical new PCs picked up a couple of USB ports almost overnight, well before the drivers were ready or there was much of anything worthwhile to plug into them, and before Apple went all-in on USB with the imacs.

Serious PC adoption was far more meandering; "legacy free" PCs were a Microsoft aspirational project/limited volume specialty option for certain cheap corporate bulk PCs for some years after the imacs went USB; and the PC OEMs were, justifiably, cautious about being too pushy about killing legacy ports before customers were good and ready; but USB ports themselves actually appeared and spread extremely rapidly once Intel decided to make it so.

What I'd like to see (and MS doesn't have; though so far as I know nobody else, at least not anyone of even remotely comparable scale) is a way for admins to 'shim' certain Oauth grant permissions.

Right now the user experience is basically "Do you want the shiny thing? Give it what it wants." The admin UI gets an actual breakdown(at least to the level of granularity that MS applies to msgraph permissions); but each one is still take it or leave it(with some specific exceptions based on backend; like "Calendars.Read" quietly being further scopable with Exchange RBAC; both most other Things.Read permissions not being). If an application has the appropriate permissions; an msgraph API query will return 'The Truth', same one the user would see, the admin would see, and any other appropriately permissioned application would see.

Compare to something like what they have for constructing SAML assertions(and what someone like Okta has a massively more intricate version of; since their whole business is being able to mash more or less any horrible combination of IdPs and relying parties together into a login flow that will work): you can just pass the AAD UPN across as the username; but you can also use other user properties and some limited text munging options; most commonly to support cases where usernames in that awful webapp are firstname_lastname or firstname.lastname@secondarydomainnamenotUPNone.org or whatever it happnes to be.

I'd like to see something similar for permission grants that gives you the ability to lie rather than deny ('lie' is an ugly word; so, um, 'enables integrated business logic flows'). Lets say some application wants user calendar access. It's for scheduling or the like so Calendars.Read is legit, it might even need Calendars.ReadWrite; but Calendars.Read also lets it grab attachments from calendar events and scrape text out of calendar event bodies: Do I really want the random SaaSholes at FindAMeetingRoom.biz to be grabbing agendas and shared documents for discussions and dial-in access passwords? No, I really don't. But I also don't want the app to break because it gets a torrent of 403s when it does some boilerplate iterating across a user's calendar events.

That's where I want to lie: there's room for potentially arbitrary levels of complexity in what you might want; do you want to have "GET /users/{id | userPrincipalName}/events/{id}/attachments" just return no attachments? Do you want any attachment that has a sensitivity label to not get returned; or to be silently replaced with a syntactically valid but uninteresting boilerplate document that's just a CSV of the meeting attendees?

I'm not expecting this to be the 'basic' application approval experience; and it would absolutely be an 'enough rope to hang yourself' kind of tool; but it's really very limiting to have a vendor-provided API that 3rd party applications are targeting that the same vendor will provide unfiltered answers to assuming the permissions are set to allow it. There's no ability to map an application's requests for information to filtered or processed versions of the data in your environment.

Obviously the API is saner than having to write some custom SQL-munging integration for even the most trivial of 3rd party applications; that would suck; but because MS answers API queries honestly; and because 3rd party devs obviously target the standard m365/0365 stuff; not a bunch of arbitrary per-customer APIs, there's no way to insert any sort of filtering(whether for information denial or for information enrichment); and all these 'applications' are somebody else's server just merrily doing its thing; so it's not even like the old days of plugins and 3rd party utilities where it is pretty challenging to know what a big binary is doing; but you can at least control its filesystem permissions and its access to the internet.

Well, yes -- the lies and the exaggerations are a problem. But even if you *discount* the lies and exaggerations, they're not *all of the problem*.

I have no reason to believe this particular individual is a liar, so I'm inclined to entertain his argument as being offered in good faith. That doesn't mean I necessarily have to buy into it. I'm also allowed to have *degrees* of belief; while the gentleman has *a* point, that doesn't mean there aren't other points to make.

That's where I am on his point. I think he's absolutely right, that LLMs don't have to be a stepping stone to AGI to be useful. Nor do I doubt they *are* useful. But I don't think we fully understand the consequences of embracing them and replacing so many people with them. The dangers of thoughtless AI adoption arise in that very gap between what LLMs do and what a sound step toward AGI ought to do.

LLMs, as I understand them, generate plausible sounding responses to prompts; in fact with the enormous datasets they have been trained on, they sound plausible to a *superhuman* degree. The gap between "accurately reasoned" and "looks really plausible" is a big, serious gap. To be fair, *humans* do this too -- satisfy their bosses with plausible-sounding but not reasoned responses -- but the fact that these systems are better at bullshitting than humans isn't a good thing.

On top of this, the organizations developing these things aren't in the business of making the world a better place -- or if they are in that business, they'd rather not be. They're making a product, and to make that product attractive their models *clearly* strive to give the user an answer that he will find acceptable, which is also dangerous in a system that generates plausible but not-properly-reasoned responses. Most of them rather transparently flatter their users, which sets my teeth on edge, precisely because it is designed to manipulate my faith in responses which aren't necessarily defensible.

In the hands of people increasingly working in isolation from other humans with differing points of view, systems which don't actually reason but are superhumanly believable are extremely dangaerous in my opinion. LLMs may be the most potent agent of confirmation bias ever devised. Now I do think these dangers can be addressed and mitigated to some degree, but the question is, will they be in a race to capture a new and incalculably value market where decision-makers, both vendors and consumers, aren't necessarily focused on the welfare of humanity?

This is obviously much harder to do under controlled experimental conditions; probably more of a cohort study; but I'd be curious if the result is more of a 'you learn significantly less' or 'your existing skill degrades'.

Either way it will at least be a problem; since the current reliability of bots basically requires knowledgeable and experienced people to supervise them and know when to just give their output a look and pass it along, when to prod them on errors to try to get them fixed, and when to just do it themselves; and you only get knowledgeable and experienced people through learning and experience; which are going to be done few favors by enhanced cheating tools and automation of entry level jobs where people historically gained experience under the supervision of senior people; but it will be a shade uglier if it turns out that using senior people to herd bots actually degrades them over time rather than just causing them to not learn nearly as much as they otherwise might.

It's not like every task is a learning experience; some are already pretty well inside your skillset and that's fine at least in moderation; but if the impact of bots is to make something like writing or programming an exercise that does to your brain what heavy construction work does to your knees and spine the future of the 40+ 'knowledge worker' looks brighter than ever!

That's why I was proposing it as one of the embarrassing failure modes. If someone at the State Department gets the wrong idea about the sincerity and consistency of the policy there will hardly be anybody for Turning Point USA to invite across the Atlantic to tell us about European race suicide without getting flagged. Awkward.

Obviously a solvable problem if you've got someone who knows how to carry out the quiet part without saying it and can do some cross referencing; but even if your social media text-munging/sentiment analysis bot is actually fit for purpose, and that's an if, it's going to be a lot of fiddly corrections both for jews who aren't frothing hard right lunatics and sufficiently pale non-jews who are.

An adversary can coerce a proprietary software producer to compromise the code. That's what we're going to see here.

An adversary cannot time-travel to when a protocol was invented, and compromise the protocol. (Though I guess the NSA can come kind of close to that, by "helping" as it's being developed, w/out the time-travel part.) That's what we're not going to see here.

Ergo, proprietary apps will remain unable to provide secure messaging, but secure messaging will remain available to people who want it.

This seems like a fantastic setup for a joke about reinforcement learning in neural networks; albeit possibly one that is on us.