It's disappointing that we still have terrible "press 9 buttons at the same time while reciting from the Necronomicon and plugging in and out the device" recovery stories.
Put the firmware and config on a ** uSD card, or set up some sort of UF2 recovery mode. Then no matter how badly you bollix it up, the answer is "plug it into your computer, download the big prominent "factory image", and drag it onto the device.
If done right, you could even do stuff like "clone the SD card so you can deploy identically-configured units for every branch of your chain business."
The keyboard enthusiasts got to "personalized hardware" pretty early, so there might be some lessons there.
There's a whole continuum from "Buy an off the shelf unit" to "here's a barebones case you can slide ready made switches and caps into" to "mix and match custom PCBs and cases" to designing your own PCBs and/or cases. There are pretty clear pipelines and even some levels of tooling for "draw up the layout you want and get a bunch of files you can send to production houses"
Takeaway 1: A lot of this is grounded in economic realities. I did the full bespoke route (custom PCB, custom 3-D printed case) and figured it probably cost me about $500 all inclusive to get what I wanted, and that's honestly a lot of money for a keyboard.
AI won't solve any of the economic problems. They can't fix "the minimum PCB order is N units, so now you have a drawer full of spares you paid for". They can't make the expensive part you needed cheaper, especially if you're an individual buying quantity of 1/5/10 instead of an OEM buying reels and containers-full. They can't change the fact that a case for a large widget will be expensive to mill/3-D print/mould/etc.
Takeaway 2: Customers may have surprisingly limited imagination for bespoke gear. There are galleries (and even coffee-table books) full of exotic keyboards. But Micro Center is full of $50 interchangeable "tenkeyless with RGB lighting" boards; throwing on a random set of "custom" keycaps, and that's enough for a large part of the audience.
Will these customers want or benefit from more tools, or will it just give them rope to hang themselves on and give them an excuse to bail out of the purchase entirely? Even if you can provide them a gallery of vetted turnkey choices, there might be more choice paralysis than actual benefit.
Takeaway 3: Hardware is forever (relative to software). You have a lot of small firms and group-buy products that disappeared and now the owners can't get an exact replacement or repairs. Conversely, Unicomp can gut and rebuild a 1986 Model M with new innards in large part because they've been selling the same basic design since a 386DX/16 cost as much as a Toyota Tercel.
If your AI spawns a galaxy of 1-of-1 bespoke products, who services and supports them? That seems like it's only going to appeal to the enthusiast-hacker type who can keep them alive themselves, who is least likely to need AI help designing them. Design for disposability isn't a great look for anything but incredibly low-cost, limited-usecase items.
I'm not sure if a bigger population would really justify more jet manufacturers.
Outside of general aviation (hobbyists/business jets), the potential buyers for jet aircraft are going to trend towards natural oligopoly.
Military craft, there's basically one buyer per country, at most a handful of different decision makers behind the same basic chequebook.
Civilian craft, does the number of airlines scale with population? There are a finite number of viable routes, airport slots, and market segments, so would it actually support a hundred new airlines, or would we just end up with a Big Southwest buying 1,000 737s instead of 100?
Even if you had more airlines, would they want to diversify their fleets? Running the same few types of planes, especially if they're the same types of planes as their competitors, unlocks efficiencies. How much would it cost a random American or European airline to retrain their pilots to fly the Comac C919, and how much would they spend extra on adaptations (stocking new spares, retraining maintenance staff, retooling processes for different floorplans)? I also suspect there's a very strong CYA/risk aversion mindset that would make it very hard to sell a new player into the market.
Regional flag carriers have political excuses to support a buy-local policy at some cost, but that still only justifies a handful of manufacturers per economic hegemon.
I think we can even skip the "that may be targeted by the US government" clause.
The whole "hosted AI" business feels like like a huge violation of corporate norms on confidentiality. Businesses that would have your head for printing out a source file to reference and annotate are encouraging developers to feed in huge amounts of proprietary code and data, and incorporate changes suggested from an outside party with minimal vetting. Evidently whatever privacy policies they've been throwing at enterprise users are plated with mithril.
At some point, one of the big services is going to get popped, and it won't just be a data breach. There's too much opportunity to quietly use the system as a malware distribution hub. Every vibe-coded dashboard suddenly starts depending on some weird left-pad fork that, 12 dependencies deep, is running a keylogger or Dogecoin miner. Your payment processor suddenly starts accepting the Konami code to approve a transaction.
I think, to a degree, ergonomics are a complete package and sound can be part of it.
I tend to err on the side of loud-click switches (Alps/Matias, buckling spring, Gateron Melodic or Kailh Box switches lately) in part because I beat a series of late-1990s dome keyboards to death because I needed some sound to feel like a key was registering.
A family member has a light linear board basically because he wanted a Pride-exploded-on-his-desktop lightshow and it's very awkward for me to use.
If you use a machine with an ISA slot, you can get a card with a chip called CH375 or CH376, which deploys a USB flash drive like a normal hard disc with either a loadable driver or option BIOS ROM. You can just pull out the entire drive and mount it on a normal Windows or Linux box.
I think the below-mentioned Pocket 376 might have one soldered-on already.
> Outside of coding, almost every business case for AI doesn’t need above human intelligence, it doesn’t even need human intelligence, or half a human intelligence, a business can extract a lot of value from a machine that has a fraction of a human’s intelligence.
What the business world actually needed isn't intelligence, it's VBA with a bit of polish on it.
Yeah, people want tools to distill reports, and puff nonsense into bigger nonsense, but to a remarkable degree, this doesn't require an LLM. In fact, the alternatives might be preferrable by offering more consistency/repeatability and efficiency (I am so sick of watching a LLM babble for 5 minutes on something a regex would do in 5 seconds).
The genius of the LLM industry is that it avoids "programming anxiety" by hiding it behind a friendly-ish UI and not calling it programming. It's another in the string of innovations like "hide the file system" and "removing user programmability" so we can sell it back to you.
There's also another side of it: there's a legitimate case for places where we don't need high quality creativity.
We have a world of clip art and stock photos. A lot of this is not best effort-- people grab what's in the catalog, and the catalog is itself often either mediocre or overused to the point where it's a blatant signal of low-effort. If you want a smiling washing machine to put on a flyer for the building's laundry room, or a generic photo of St. Basil's for an article about Russia, nobody really complains about that.
AI could fill a similar niche. We know it's crap, but it's good enough crap. Except it's absurdly expensive crap when the externalities are actually priced in.
For some of us, we found digital "third spaces". In a way, when we're arguing on HN, or the kids on Discord, are using it largely as a community centre on their screen.
It clearly shouldn't be the only option-- it's not going to service all types of interests and audiences. I attend the meetings of the local numsimatics club-- they get a room in one of the city offices when it's closed for the day, and do presentations and discussions. The demographics are weird, it's like under-15s and over-60s with very little in the middle, and I'm not sure there's the technical capability or the preference to do it over Zoom. (Trying to remember what they did during the pandemic, I think they just closed down for a year or so)
Isn't the solution just to take the market out of real estate?
At the individual-homeowner level, a lot of NIMBYism, HOA activity, and other toxic activity is grounded in "I have $400k of equity sunk into this square of land and it's the only meaningful asset I have for my increasingly oncoming retirement."
There are probably plenty of people who are subconsciously self-sabotaging due to that-- they may personally prefer a denser neighbourhood or more mixed-use facilities, but that Zillow number is blinking in their mind at the same time.
If the real estate were nationalized and people were leasing it from a state-owned trust, at basically its cost of operation (maintenance and amortized construction costs), it would free up a lot of resident income for other use or investment opportunities, and make a lot of changes less threatening to residents. (ISTR claims that rents in the USSR were on the order of 5% of income, and this when 30% or less was still a feasible story in the US)
We lost this war when we got bamboozled into putting the V-Chip into televisions.
The obvious tradeoff was that we should have been able to have all forms of offensive and pornographic choices on the public airwaves, because we've given those who are concerned the tools to explicitly block it. (not that "unplugging the set when the parents weren't around" wasn't a viable tool already).
We never got that.
I do wonder how much of it is directly that the "won't someone think of the children" demographic is politically loud and courtable in and of itself, and how much of it has been fostered by firms that see it as a conduit for more nefarious aims (i. e. commercial social providers who want desperately for a legal CYA so they don't have to do the dance of COPPA compliance and have an incentive in the verified demographic details age attestation provides)
I'd suspect one of the big constraints was that you had to support basic VGA: 640x480 at 16 bold colours on a cheap CRT.
With that palette, using white/black/dark grey/light grey for the UI leaves most of the "saturated" colours for the software's own messaging. If you make the main body of the UI red or yellow, it's going to wash out if software uses those colours for alert indicators. If you make it blue, it's probably going to be a bit subjectively dark, and you'd need to use light characters for contrast. White input fields might have a skeumorphic aspect in that they look like paper you can write on.
An interesting comparison is the colour schemes of CDE. It was usually used on 256+ colour systems, so many of the themes are pastel-centric, but it's aggressively beveled and has similar high legibility.
I can't quite understand the "fixing small errors is easier when just talking to Claude" flow.
I tried having it write some tests today. It got very close to what I want, but picked a stupid set of input values (two fields that look independent that should only be used with related values). I thought about "how do I explain this" and then just went in and fixed it myself.
How is it easier to write "Okay, go back to testBlah and change xxx to yyy" versus clicking on XXX in the IDE and typing YYY by hand? Maybe if you had 500 faulty tests and were forbidden from using search-and-replace for some reason.
It makes sense when code generation is the limiting factor, but I end up with a lot of changes where the actual code delta is smaller than the necessary prompt to convince the bot to produce it.
My point is that you can't cherry pick a profitable business unit if you don't have a story for how the entire business can operate profitably.
Cars have low margins, but they generally don't rely on an ongoing infusion of investor money to balance the books. The overall venture still has to turn a profit. Nobody is walking into Hyundai HQ and saying "We are going to sell Sonatas for $6,750 each, because if we can do it long enough, some magic will happen and we'll end up back in the green."
TBH, I'm not quite sure what the "some magic will happen" angle is for AI.
Compute gets cheaper, but I suspect the training arms race is running up costs faster still.
We're already seeing hints of price balking (definitely heard people at work saying they're hesitant about Fable due to the costs) so it's unclear if there's headroom there.
TBH, the best answer I can figure right now is that many players are hoping for a competitor's flameout-- the dream of being the last man standing, and then able to dictate market terms.
In the coin collecting universe, which has had a world of third party grading for close to 50 years now, there's a saying "buy the coin, not the holder". But at the same time, there's also a sub-market of people explicitly buying the holder. People will pay a comical premium to hold the "finest ever graded" (or conversely, the "lowest ever graded") example of the same basic date-and-variety.
The difference is that the coin market evolved this ecosystem somewhat organically. The services originally started more towards "proving a rare date prone to counterfeiting is authentic" and "providing a neutral third-party since grading can be imprecise and subjective". People didn't start considering the holder for quite a while.
I think part of the ickniness is that this smells like it was designed to puff up the legitimacy of the grading service as much as anything. It's not like taking something that's understood as broadly rare-- some one-off prototype or widely known scarce item (like the World Championship cartridges). Instead, they're taking something very common and trying to focus on "peripheral" rarity factors-- a sub-variety (some particular early packaging) and a condition rarity (most units are no longer untouched and sealed). This has a smell of "you too could turn a fairly typical item into something life-changing by sending it in for professional certification!"
We get that stuff in the coin collecting world, but it usually doesn't make as much hay as you'd expect. People still get way more excited about the simple "1804 dollar" than the probably rarer "1817/4 half dollar, Overton 102 variety, top grade"
On the one hand, it's nice that Epson provided a Linux package.
On the other, it's a pain in the arse that it's specific to a handful of distributions.
I had a V33, and I use Void. At one point I was able to get it shimmed together, there was a wrapper package for the vendor-specific modules, but eventually it stopped working, so I went to the Goodwill and came home with a Canon LIDE 100 that Just Worked with SANE.
I'm not sure what I'd do if I had to buy new. It feels like the entire consumer scanner market is sort of dead. I can recall when you could go into any computer/electronic big-box shop and see a bunch of cheap little flatbed parallel (or eventually USB) scanners for like USD50 or less. They were a common sweetener in the name-brand PC bundles of the late 1990s/early 2000s. Now there's virtually nothing under USD100. It feels like it's a solved problem, too-- most home users don't need ever-increasing resolution or scan speed, so they could just keep cranking out 20 year old already-amortized designs.
"Profitable on inference". Isn't that exactly the same a physical business saying "our widgets have a marginal cost of 95 cents to make, and we can sell for a dollar, so we're profitable, as long as you forget we have a $92 kajillion loan on the factory that has to be serviced."
It raises a chain of interesting questions: what if we pulled the plug on the expensive part (the training and associated infrastructure) in pursuit of making it economically viable?
- How much of the audience is using it based on the long-term promise? "It's still imperfect and annoying, but I want to be ready for when it finally turns into Lieutenant Commander Data." If the vendors said "this is what you actually get once the honeymoon ends", would customers still be satisfied with the product and pricing?
- How do you stop the game of economic chicken? If Anthropic said "Fable is the last model we can offer (until we can pay down the costs to get there)", any competitor with a dime of runway left, will spend a cent of it on training and 9 cents advertising "do you want to be stuck with old tech?"
Sometimes you want the machine to be an advisor and sanity check your suggestions.
Sometimes you just want it to do the boilerplate you have in mind without trying to reason everything from first principles.
I told you to check fields "foo" and "bar" for values "baz" and "quux". You don't need to go diving through the entire source tree to discover where and how this is set.
I guess maybe it's helpful for the vibe-coded audience-- if it tries to over-process everything, there's a better chance it will work on a single shot, but I'm taking the Crazy Taxi approach: you get points if you drop me off within 20 metres of where I wanted to go, and I can correct it if I specified the wrong response message in the original approach.
Put the firmware and config on a ** uSD card, or set up some sort of UF2 recovery mode. Then no matter how badly you bollix it up, the answer is "plug it into your computer, download the big prominent "factory image", and drag it onto the device.
If done right, you could even do stuff like "clone the SD card so you can deploy identically-configured units for every branch of your chain business."