I remember the Internet before Eternal September. You need to go back at least that far.

Not a financial person either, but wouldn't writing it down mean that a company no longer considers it to have value? Is there something unique to the hardware they use for search-specific AI that prevents it from being leased out as part of Google Cloud or being used to run other teams' models?

Not like IPv6 solves anything with lots of ISP blocking inbound IPv6 connections.

Free speech has never been free as in beer, or at least not at any meaningful volume level. Talking to a single person was free, but that's not going to spread your ideas very far or very quickly. In the early days, you paid scribes, but that would cost you money. Later, the printing press brought the costs down, but you still needed money. Radio and TV made it possible to reach even more people more quickly, but at a significantly nonzero cost.

The Internet is no different in that regard. In theory, you can always spin up your own server and post anything you want to. But that won't guarantee that anybody is going to actually visit your site. To get eyes, you'll still have to buy ads, etc.

Normal industry: Your factory has an expensive milling machine and a cheap stamping machine. Thankfully, customers actually seem to prefer stamped parts for most cases and you can produce those quickly; so you keep stamping out parts and rolling in cash.

Tech industry: As above but your entire c-suite believes that the milling machine with turn into mecha-jesus and end history; so you are bulldozing everything you can to build more milling machines.

Right that is the real problem isn't. Google has not yet worked exactly how to stuff enough ad impressions to make money without destroying the utility the user is getting out of the experience to the point they are likely to abandon Google for a traditional search engine or go pay for Chat-GPT or something.

Really 'ad words' in AI summaries ought to be "easy" from an implementation standpoint. The 'problem' is doing its going to use even more tokens that need to be paid for.

I agree

The winners here are going to be people with actual hardware IP (Nvidia, Apple, others), people with actual hardware fabs, and likely Google and Meta. I predict despite the current crazy valuations the Magnificent-7 eventually end up acquiring the leading AI labs.

The real costs are in building models, not hosting them comparatively speaking and that picture only improves as the energy infrastructure and hardware markets gradually align to the volume demands. The labs are already just reselling the hyperscallers computer, there is no reason those guys are not going to take 'good enough' models and beat them on price.

Having the 'best' models is an asset someone with existing SaaS, Marketing, etc business on the hand can leverage and by extension justify the expense. That is why I see OpenAI and Anthropic eventually becoming acquisition targets once the gold-rush investors capitulate on their valuations.

The option is for them to get directly into the datacenter biz (not contracting for them and agreeing consumption commitments but actually building them and owning the dirt all they way up to the roof top chillers. Anthropic seems to be trying to make that play but I still don't know where the revenue will really come from...

Yes, but not for you and me. Graphics designers have been using stylus-displays for years and years, because it's great to draw on the thing you are making instead of drawing on a different surface like a Wacom.

But I'm like you, I don't have a use case for a touchscreen and if you touch my notebook screen I'll be angry because I prefer it without fingerprints, thanks. So my first question is: Do they offer this thing with OLED but without touch? I guess not. Pity.

I've been slowly moving towards my dream setup: Solar panels, battery, EV car, island mode capability. This is one thing that'd be a factor in my buying decision. Using your car as a battery or secondary battery is a big plus, and if they built the necessary equipment to run independent of the power grid, it'd come in a neat package.

While power failures are rare where I live, I expect them to become more common with the move to renewable energies. I support that, but it's more fragile than constantly running power stations. So being able to run your home at least for a few hours on your own energy is something that I consider a good idea.

It seems likely that the answer to this question is 'yes'; but it seems like a mistake to be distracted by it when we are talking about people who are, busily and intentionally, building 'AI' products against the public.

It's mildly comical when they attempt to emulate humanity; but the important bit isn't the cluelessness; it's the malice.

I suspect that this would be more tedious(and potentially less flattering); but you'd think that the metric you'd really care about for the juche-computer would be component availability.

Going by cost is interesting if you are trying to get the attention of balance-of-trade types and tariffs and dumping hawks; but it's an almost perfectly wrong measure if you are trying to catch any "oops, turns out that only the Chinese vendors so hungry they are considering offshoring to Vietnam still make a random passive or unexciting tolerances switching IC because it's a zero-margin part with no real room for differentiation that everyone who could got out of years ago" type situations.

If it's a real, drop-dead, 'national security' problem some of those are probably less serious, if there's a much fancier TI part that is overkill for the job but could be subbed in for only 3-4 factors of ten more; but if you want to keep economically viable civilian production open or it doesn't have a direct upmarket analog that could get really awkward; even if you've ensured that all the fancy logic ICs, copper, and gold are geopolitically friendly.

That's basically the Nexperia story, slightly simplified. Philips decided to private equity semiconductors generally; NXP decided to shove a bunch of the boring parts out to WingTech; then 'oops' European manufacturers who need diodes and ESD protection parts they thought were cheap jellybean stuff have cars piling up on the line because they can't get them.

Obviously losing a really cutting edge capability isn't preferred; but it's also the sort of thing that people notice because those tend to be pretty valuable; while low and middling capabilities, including ones that are sophisticated enough that you can't just throw money at restoring them fast, are things that are voluntarily surrendered because getting rid of low margin products makes line go up.

You missed the point.

A friend of mine made a tea-timer app for the iPhone. Where you enter what tea you are making and it sets the correct number of minutes. Today, people would ask ChatGPT before they go on the App Store to search for something like that. That's the point. Not what YOU would do, but what the general public does. And for Jane Doe, ChatGPT has replaced a lot of these specialty apps.

Skills are great until they aren't. The problem is exactly what you say: You can literally create a skill by telling the LLM to create a skill. Which is... why the f do I need to make a skill if the AI already knows how to do it?

Properly created skills, that include reference material, specific instructions, and expert knowledge the LLM doesn't have or that's buried somewhere under 10 layers of reddit posts is genuinely useful. But most skills aren't. I'm constantly underwhelmed by hyped-up skills that, when I look at them, turn out to be bland, basic instructions.

Only in the most utterly pedantic sense of the word "different" wherein you mean that by definition, infant mortality occurs early and wear occurs later. But they can both stem from the same root cause, which is a component not built with adequate tolerances. In other words, they can be the same fundamental issue, just expressing themselves at different times.

Most products don't have an MTBF. And things that do have MTBF are based on accelerated wear testing on a small number of units, and can have almost zero basis in reality. There was one model of major brand hard drive about twenty years back where a couple of RAIDs where I used to work got lost because the drives all died after such a predictable number of powered-on hours that by the time you could re-stripe the RAID, the next drive would die. They ended up making sure the drives had a high variance in powered-on hours before using them in RAIDs after that.

It's an example that proves or demonstrates a point. And the point it was proving was that the same root cause can be responsible for both catastrophic early product failures and premature deterioration-related failures.

Wear-out failures aren't always different from infant mortality failures, though. If the defect is caused by a design flaw in a component resulting in a structural weak point, that same weak point could fail after one use or a thousand, and it could just be a ticking time bomb. Someone could get very unlucky and find one that is slightly weaker, and it could break much sooner.

Case in point, I recall a machine where some people would have to have their laptops replaced every couple of months for motherboard failures, while others lasted for years, and it turned out to be dependent on how people closed their laptops at the end of the day and whether they pressed on the bottom of the case in a particular spot that wiggled the GPU. But if you handed one that had been working for five years to someone who used their left hand under the laptop while closing it, it would be dead in two or three months. If you replaced it with a brand new one, the brand new one would be dead in two or three months. And so on.