hakfoo

I suspect there's a place for LLMs, but it would be one that gets developed on a per-organization or even per-team basis, rather than the tendency to airdrop some elaborate do-all solution.

This is probably why it's chafing on a lot of developers. We're hearing we have to use these tools and only these tools even when they aren't great fits for specific tasks or workflows.

The management loves the idea it closed an inbound ticket in 30 minutes... by converting it to a code change that fails to fix the actual problem, or by replacing it with a slightly different category of ticket that becomes a burden for someone else to finish off, which is for some reason not a metric anyone cares about.

Consoles are a struggling market right now.

The architectural differences that made a console more compelling than a standalone PC are largely gone. (Nintendo is a bit of an exception here, but they were always playing a different market than Sony and MS anyway)

This means it's harder and harder to say "here's a unique experience you only get on Xbox" (or Playstation for that matter)

The price advantage is going away fast, especially with talk that next-gen consoles will top USD 1000.

All they have left is their prestige first-party products, which they don't seem to really know how to use effectively. Some of them have solid cultural penetration, and could spawn broader media franchises with the right handling.

Maybe the problem is libraries themselves.

All else equal, I'd rather have the richest possible functionality built into the language. At least I know that it will be updated and maintained in one place, and there's very little ambiguity about setup. You don't have to start picking apart the fact IN THIS PARTICULAR PROJECT, libfoo was imported by pasting it in the code base in 1977 before package management existed and globally namespaced, but libbar is a first-class citizen in the depency-injection system.

If you know you can just call bargraph($data), even if the stock bargraph library is boring, then people and LLMs will favour it.

I've heard the argument that a lot of the 'scarcity' in RF spectrum comes from chicken-egg problems based on the original allocation strategies.

If you give out a dedicated fixed slice of the spectrum, and high power licenses are pretty limited, you end up treating it as a "circuit switched" resource. I have my narrow sliver at 144.4449MHz and I will use it as I see fit.

If you allocate capacity with the understanding it will be shared, it creates a system where you have to think more "packet switched" -- how do I detect congestion, retry, find less congested workspaces?

The "packet switched" philosophy usually supports a lot more users on the same physical infrastructure

Cloud wasn't sold as a performance thing. It's more about outsourcing the maintenance (let AWS replace a dead PSU at 4AM) or for the promise of scaling tasks to infinity for dynamic demand (web hosting).

You might see interest in thin-client plays for the same reason. They'd like to replace the local desktop with throwaway $50 terminals because it's easier to deploy on a standardized VM image and not have to replace dead SSDs and CPU fans.

Workstation tasks that require low-latency and high performance seem to stay local because the performance is simply not there.

That's the second question I raised.

As you said, you still have 1/8 or 1/10 of the overall costs in OpEx to cover.

You still have a lot of services of such marginal utility that they'd have a hard time justifying their opex costs alone.

I think part of this is that we have a lot of firms offering features rather than products. I fully expect most of the third-party "meeting minder" and "writing improver" services will be replaced by versions built into the platforms themselves, for example.

> Even if the investors funding this construction all go bankrupt, the data centers don't just vanish: someone will buy them from the bankruptcy court and operate them.

There are really three questions to answer on the long term:

1. Is an arbitrary service profitable enough to cover opex AND burn down the massive capex financing? A few companies might have enough magic to survive, but I suspect a lot of them go to the bankruptcy fire sale you mention.

2. Is the service profitable enough to cover just opex? The second generation vendors may have a somewhat more favourable cost structure, but there's still a floor. Electricity and cooling on a datacentre scale aren't free, and we may find there are still services that can't even meet that threshold. I notice you see far less marketing for consumer video-gen these days...

3. We let those businesses fail. What does the long term market look like at that point? You have a snapshot in time of paid-for datacentres, but the market shock probably shot a huge torpedo into long-term R&D and planning. Vendors will scale back to refocus on markets that are more vital (maybe selling RAM and GPUs back to end users :) ), and create an era of stagnation. AI becomes a utility, in the sense of "boring, predictable industry that growth investors have very little enthusiasm for".

There is also the risk that the data centres could vanish. This is a very hype-dependent market: we've spent the last 5 years going from "novelty" to "productivity tool" to "weapon of mass destruction" to keep people excited.

Eventually there's a put-up-or-shut-up moment. What happens if we get to 2035, we've hit a plateau for 3 or 4 years, and they can't deliver Lieutenant Commander Data, or even Kryten? All we really have to show for the effort is a bunch of old GPUs, displaced workers, and the ugliest restaurant fliers you've ever seen. It's a big loud target for a bunch of frustrated and let-down people. We're already seeing moratoria getting political legs, could permit revocations or forced decommissioning be far behind?

I suspect this is part of the reason some players are happy enough to present AI leadership as a New Cold War/US v China thing-- pull the national security card to try to shut down a debate on the actual utility and desirability of the actual product.

The HFT industry creates an investment opportunity that's somewhat detached from business fundamentals. If you can trade on the volatility alone, you no longer need to pick investments based on their future prospects and behaviour.

It didn't help that people are seemingly more offended about the possibility of spam on their inbox than on losing a valuable email due to misclassification.

There's little to no incentive for providers to ever de-escalate.

At my day job, we use Gsuite. I never used it for any non-work purpose, and it wasn't publically published, so any "spam" it got was false positives. Emails I actually needed were being buried below the fold in the spam folder. It turns out there is no obvious way to 100% disable spam filtering (at the time I got the first false-positive at least) so I had to try to just allowlist things that should be left alone.

It tells a specific counterfactual story, but it's one that has a reasonable background for it.

If you were a new computer user in 1980, your Appie II+ or PET 8032 booted straight into BASIC, and it was a good-enough tool that could do productive development for the entire life of the machine there.

But if you were a new computer user in 1990, you didn't have that. They gave you GW-BASIC, and if you're lucky a separate manual rather than a vague chapter in the main DOS manual. This might be enough to give you some initial experience with programming, but you have a bunch of hard brick walls.

Your machine might have 640k-4Mb of memory, but GW-BASIC can only meaningfully work with 64k of it. You can't access the mouse or any post-EGA graphics modes without a LOT of assembly shimming. If you had a sound card, it was outside the built-in tooling.

Remember, this is a few years before the Internet being widely available, and stuff like the GNU project being widely accessible. This is probably the best development tool you have available

If you wanted to go further in programming, it meant buying a boxed compiler that probably cost well in excess of USD100, and was for a completely different language than the one you had cut your teeth on.

But if you wanted "The GW-BASIC I already understand, but more", you want this.

The play that finally sunk my old job (headed by a serial entrepeneur) was literally selling a $49 product with a $80-per-transaction commission.

We were supposed to make it all up on being able to hard-sell the user on addons and multi-pack buys.

We're one well-publicised security pop from locally hosted being a non-negotiable for some large set of buyers.

They've convinced every company in the country, including in previously security or compliance-centric environments (finance, medical, government), to hand them hyper-sensitive data.

We've already seen the leakage scenario, with the academics who found the labs scooping them on the research they were using the LLMs to assist. What's to say they-- or someone who infiltrated their infrastructure-- won't do the same thing with your propriatery business details?

I'm not sure open LLMs are a great play for commodity vendors. A lot of "lowest bidder commodity services" work because they can wildly oversell. That VPS hosting a small-biz website or 1000-messages-a-week internal email can run on a Raspberry Pi 1B, so you can pretty easily sublet a $1000 server to 50 customers at $10 per month each. Current LLM hardware plays tend to need more expensive gear and seem to be running at higher utilization rates, so the price that actually fits the underlying business might not look as appealing.

As I see it, the value in a wealth tax isn't just in a cash grab. It's about reining in the behaviour of the wealthy.

If you can cap individual wealth to eight or maybe nine figures, you take options like "bankroll major state/national elections on a whim" and "buy up major media and communications brands" off the table. As far as I'm concerned, these are things we can do without and probably have a healthier society as a result.

California's attempt might be well-intentioned, but it seems so weaksauce that it won't reach to the behaviour-modifying level.

The "why bother investing" mentality that's used to argue against Georgist land-taxes might be interesting applied to a high wealth tax too. If your stock portfolio is already approaching the cutoff line, will you press as aggressively for that last short-term bubble with long-term expenses?

It probably makes sense with a silent linear keyswitch for users who are used to the obvious acoustic feedback of a typewriter.

Since cheap modern MCUs have programmable ADCs, you can wiggle an output pin and experiment with different sounds. I've tried something a bit more "tinkle" like than a flat "beep" and it has its charms, but it could go a whole new place like different tones for special keys (simulating the clunk of a shift or space bar) or even keys that you could subtly differentiate by sound to tell if you're off alignment.

There's probably a better take on this in the "non-sold" software market. I feel like there's an optimistic story there, in that AI could lead back towards an era of user-directed computing.

We've had a long trend of software vendors infantilizing and limiting their users, normally in the service of some goal of making the product nonintimidating to nontechnical customers (mobile-first design with hidden or nonexistent poweruser functions), or to simplify their own job (Electron instead of native)

But on any system people use frequently, users eventually want to refine and streamline their workflows, to maximize productivity and limit frustration.

I suspect the slopamatron will be the 2026 answer. 20 years ago, you might have non-developers scraping together just enough VBA, TCL/Tk, AppleScript, or even macro-recorder tools to file off their rough edges. Today, those users will ask the bot for help to wrestle back control, particular when a lot of the old hardpoints for automation don't exist in newer designs.

Most of these tools never end up on an App Store-- at best they'd be shared internally within a single organization, or a user-group if those still exist, but they're delivering way more value than the a random App Store game with 7 downloads and a bunch of legally distinct knockoff assets.

Maybe if vendors see that large numbers of their customers are enlisting AI to work around bad design decisions, they'll reconsider those decisions, or at least surface first-class automation tooling, which could be consumed by chatbots and human programmers.

This seems like a new spin on the "wealth fades after 3 generations" narrative.

You built the killbots that prevent anyone from ever reallocating your wealth and holed up in your compound. But will your daughter and granddaughter be able to keep them operational and aligned to their interests?

Phone batteries have relatively little margin. If it lasts barely a full day at full capacity, going down to 80% means you can no longer reliably say "I'll charge it overnight".

A car battery that lasted 300km new but now only 200km is still entirely adequate for a commuter with a 25km typical commute, and occasional 75km days out.

Western manufacturers really shot themselves in the crotch with this one. Foot isn't strong enough here.

Had CXMT hit a normal functioning market, they would be a much harder sell-- existing vendors could spread FUD about performance and quality to hold them back for a few years. But now you have an audience that is desperate to the verge of soldering up banks of 74LS74 flip-flops to make their own RAM.

It's not even like they're offering a reasonable "there's a domestic product but it's marginally more expensive" alternative. And it's all their fault. I sort of wonder if someone was actually saying "we'd better keep some supply around just to avoid burning our bridges with Dell/HP/Apple/etc." when the deals went down.

The US might try to go protectionist/China-baiting to keep CXMT products off their market, but I expect they'll sell just fine everywhere else.

The billing address "bull" is part of a basic fraud due-diligence feature (AVS - Address Verification System).

You can get back different match levels (matched the house number only, the zip code only, or both) which provide some insight into potentially sketchy transactions. Some payment gateways allow you to auto-decline based on the results, or the merchant can choose to follow up depending on their scale and risk tolerance.

Plenty of petrol stations ask for the zip code on the unattended keypad when making a purchase for the same reason.

"Donate to a different charity" is in no way a viable alternative.

Charities are anything but interchangeable. You might not want to sponsor one on ideological grounds, or concerns about how effectively they use the money, or just because you were seeking to support a specific cause. If you deposited money in a box that said "we're buying medicine for sick people in Somaliland" and they sent it to the RSPCA instead, wouldn't you be pissed?

"probably still requires a lot of GPU compute" is something that would be very desirable to undercut.

There's a galaxy of potential AI markets that are local-only for compliance/privacy/deployment environment reasons, and an equally large and overlapping set of use cases where you won't be able to bolt a rack of thousand-watt GPUs on the side of the device.

What if the next "DeepSeek shock" is something we can run for non-toy use cases on our box of old Android phones? The GPU data centres could be expensive albatrosses very quickly.

There's probably a case that right now, the bigger-is-better paradigm protects incumbents-- a smaller model will always have a FOMO factor unless it can be proven competitive, so that leaves the playing field to those who can afford to train and deploy a new Fable or Sol every few months.

I guess I don't understand the sales argument for a 'smart monitor'.

For the "smart TV" the argument to the consumer is either 1) to access OTT services if you don't have a device for it already or 2) that the data harvest subsidizes the product.

Neither of these arguments make sense for a straight monitor.

People generally don't buy monitors if they don't have a PC, game console, or some other video source to plug them into. So "it runs Netflix or Zoom standalone" isn't gonna move the needle.

The subsidy argument is also pretty hard to make. It works for TVs because you can advertise "60 inches with malware for the same price as 32 inches without". But monitors have fairly low upper limits on desired size when you're using them at a desk. Many would find a 60" screen overwhelming, but 27" is reasonable. You can buy a new 27" 1080p IPS monitor for $95 today. Can ads and data theft subsidize it down to $80? Maybe. But again, will that make much of a difference for consumers?

This is also only from a consumer perspective. I can imagine IT department having a fit about these products. At the best, it's support tickets when someone accidentally engages with an unexpected feature or plugs their Ethernet cord into the monitor. At worst, it's a fun new source of compliance obligations and patch processes.

The worst-case angle I can see is eventually media entities will see it as a way to close one more loophole. Someone will launch a service that can't be streamed on the host PC, only on blessed smart devices themselves, to eliminate the risk someone plugged the DisplayPort cable into some sort of DRM-stripping device. But that would alienate a hell of a lot of potential customers.

Different arguments for different audiences.

Some people respond to moral arguments directly.

But even if you're all about extractive imperialism, you can make a case for "starting fires in your own backyard is going to be costly for you, especially in the long term."

It's interesting to see their intent to make an explicit drop-in replacement.

I'm surprised there aren't really high-quality drop-ins for most of the classic "name brand" commercial fonts at this point, built solely out of the commercial motivation. "We could contract out someone to build it for less than we're paying a commercial foundry to license it for N years, and then forever every designer is using BrandCo Sans instead of Futura/Univers/etc so we get the free brand recognition".

Bread and Circuses was the old answer to social unrest.

The new bet is on surveiliance and security tech, the more autonomous the better. The owner class no longer needs to provide bread and circuses when they can constantly monitor for any sign of dissent and deploy enforcers without empathy or morals.

Property rights might not end up as sacrosanct as they hope. There's a series of slow-motion emergencies unfolding, and emergencies are often a place where all rights-- not just property rights-- get curtailed. If they keep up the "Let them eat cake" act, it's going to make them an easy target for a gritty reboot of 1917.

It's a very pretty benchmark.

Everyone has the same basic objective and measurement norm, so we can see how performance improves year over year, and how different manufacturers handle a specific known goal, even if it's not super-representative of real world use cases.

Billion-dollar supercomputers don't spend most of their time running LINPACK, and some probably perform better on specific tasks, but it doesn't make the Top 500 list any less of a thing.

There's a good chance that most humanoid robots are not going to look particularly hunan.

You might have different proportions due to different weight distribution or use cases (maybe the legs are made less flexible if the workspace has fewer obstables, for example.

Maybe "bipedal" is a better descriptor, since wheels are a proven way to go faster than Usain Bolt, but walking-style strides are new.

The hard part is knowing where to draw the line between relatively stable infrastructure items that don't benefit from duplication, and flexible items that are likely to benefit from specialization and competition.

X11 made the right choice in putting graphics drivers in the "stable infrastructure" stuff and display toolkits and window managers in the "specialize and compete" section.

I'm sort of pleased about that we're finally trying to replicate this division with the River ecosystem, building a standard compositor that we can plug in different actual window management logic into it, but that makes River itself a huge single point of failure-- if it has a bum release or fails to keep up with trends and extensions, it takes down a lot of the ecosystem, but perhaps not enough to be a five-alarm fire the way Xorg or XF86 back in the day would have been.

I've always figured the dream would be to be able to advertise a low price without ever having to deliver it.

Some particularly skeevy car dealers were skewered many years ago for saying "Advertised price is less tax, title, and additional $2,000".

But they CAN get away with saying "The $8.99 burger is $11.99 unless you: - possess one of a range of sanctioned devices - have access to data services - are able to create and manage an account - are willing to opt in to whatever tracking/marketing gimmickry - can claim the "deal" in advance

Each of those steps is a funnel that opts out at least some potential customers from the deal. The 12-year-old who doesn't own a phone can't get the deal. The 85-year-old who can't operate Android can't. The person making impulse purchases and not pre-researching their meal selection misses the offer.

I feel like there's a argument for a precedent akin to "no purchase necessary to enter" for sweepstakes-- no additional purchase, software install, or commitment should be required to claim any advertised offer.

A big part of the numbers game is on variants though. There's what, eight or ten total RP2040/2350/2354 parts to choose from? So all the volume is split between those.

Conversely, there are a lot more variations within other MCU lineups. Just looking at the WCH CH32v30x datasheet-- a single relatively low-popularity range from a single vendor, the datasheet shows 9 basic models with five different physical form factors, and there are theoretically different SKUs for different temperature ranges, etc.