Animats

So how is the part feeder setup automated? That's the big problem for short runs. Board making, solder paste, pick and place, reflow, and inspection were all automated already. The usual short run setup bottleneck is that someone has to put the tapes of parts into the feeders, then load the feeders onto the pick and place machine.

The video doesn't show how that happens. The robotic systems seem to be moving boards around, not parts. Yet that's the big manual part.

EHang keeps demoing their flying cars, without getting much attention. [1] They're scaled-up quadcopters. Their newest model transitions to winged flight to get better range. They've been testing versions with expensive solid-state batteries to increase the range.

Archer Aviation has been flying their electric air taxi over the SF bay this week.[2] Cost is about $5 million, though.

The big event for flying cars will be the 2028 Olympics in LA. VIPs will be shuttled around in Archer craft. That's about the right demo market - trips are short, but congestion on the ground is a problem.

[1] https://www.ehang.com/

[2] https://www.archer.com/news/archer-completes-historic-electr...

Right. The S.S. United States was built for winter North Atlantic conditions, with a power plant, hull, and watertight compartments more like a warship than a cruise ship. She could be converted to a troop carrier if necessary, something that had been done to big liners during WWII. Cruising speed 30MPH, top speed 44MPH. Great for getting across the Atlantic in 3 1/2 days. Not so great on fuel economy.

Cruise ships today look like hotels mounted on barges [1]. About half the speed. A refurbished SS United States was never going to compete in that market, and plans to do so were never funded.

[1] https://www.royalcaribbean.com/cruise-ships/icon-of-the-seas

> Zeppelins, though. Those, I will grant you, have gotten scarce.

Just saw a Zeppelin cruising over Woodside today. It's Sergey Brin's personal project, Pathfinder. It's a true Zeppelin, a rigid airship, not a blimp, and the world's largest air vehicle. [1] First flew in 2025, but, unlike the Airship Ventures craft we used to see giving tours over Silicon Valley, it's not flown often. It's normally parked in Hangar One at Moffet Field.

[1] https://www.kron4.com/news/bay-area/big-helium-airship-spott...

And yet Amazon has Alexa. They have a user base for a smart chatbot. What's Alexa going to run on for hard questions?

Full startup and shutdown of a coal plant can take days, but throttling down to about half power during low-load periods is normal. Power plants do this every day. Late at night, load decreases and plants throttle down. Every morning, between 0600 and 0900 local time, there's "morning load take-up" as load increases.

The big issue is what's called "ramp rate" - how fast can output power be increased or decreased? This tends to be around 1% to 3% of capacity per minute. See Figure B-2 in [1], which is a detailed discussion of how to improve your coal-fired power plant's ramp rate.

Natural gas plants have much higher ramp rates, and for batteries, it's under a second. Nuclear plants can be ramped up and down, slowly, but since it's all capital cost, cutting output raises the price of the power.

[1] https://www.epri.com/research/products/3002012979

It looks like Energy Vault, the people with the tower crane and big stack of concrete blocks scheme, gave up on that. Energy Vault is now selling large battery systems that actually work.

Oh, Advanced Photonics Coalition, not American Power Conversion, the UPS people.

Most 3-letter acronyms have been used. If you're naming a new thing, check first. Use 4 or 5 letters if necessary to get uniqueness.

Not by this result. But it opens up a new area of problems to hammer on that might get somewhere.

That's true at the beginning of a new field. I'm noted for some advances in network architecture. But when I did that, everybody working on the Internet could meet in one large classroom. If you were reasonably good, you could make major progress.

Computer science used to be tiny. In the late 1970s, AI was a few tens of people at MIT, at Stanford, and at CMU, with a few smaller groups elsewhere. I met about half the greats of my generation, as did everyone else who was serious. There were almost no textbooks in the early 1970s. When Knuth's Fundamental Algorithms came out, it was great to finally see what had previously been folklore on mimeographed sheets in an actual book.

I was involved with Autodesk in its very early days. Even went to the founding meeting. The feeling was that PCs were finally getting big enough to be interesting. The founders were all mainframe programmers, and the people who came up from 4K memory on an 8088 had no idea how to organize a big program. But mainframe people did.

There was a vast emptyness in the product space ready to be filled. Products considered were a window manager ("AutoScreen"), a text editor (Kern Sibbald actually built that), a database, and a CAD program. None of those existed yet as successful products. CAD was picked because the productivity gain was so huge vs. paper and pencil drafting. At the time, the state of the art in drawing editing was drawing on mylar and using an electric eraser to delete. Then you ran the Mylar master through a big copier to make blueprints or whiteprints. A CAD system on a PC was a very easy sell.

So all you had to do was make it work. No need to overcome the noise level of a thousand similar "apps".

There are hard problems where the worst case may be exponentially hard, but the average case is much faster. Linear programming and the traveling salesman problem are like that. The faster approaches are considered "probabilistic solutions".

In cryptography, most of the problems believed to be exponentially hard, such as factoring, don't have proofs of an exponential lower bound. There's some risk of a breakthrough that makes those problems easier. A probabilistic solution would be good enough to break many cryptosystems.

I suspect that people at the cryptanalysis agencies worldwide are staying up late tonight, trying to figure out the impacts and risks of this new result.

This is a big deal. It's Reuters, not some blogger, saying this. The STAR50 index of AI and semiconductor stock in China is down 40%. P/E from the index dropped from around 250 to 100, which is still high. The investment community in China is concerned about other indices over-weighting tech stocks with high P/E ratios. Then index funds pick them up and funnel capital into them. AI stocks are 42% of China's market capitalization, so index funds are bent way out of shape.

Tesla's P/E is somewhere around 450, incidentally.

In a few years, assisted suicide will probably be offered as a feature of offboarding packages.

"Your work is done. Why wait?" - (to paraphrase George Eastman's suicide note.)

> This is no different to our parents generation that used to tell us to stop watching TV or get off the games console (like a Gameboy).

It's improved. When cell phones first came out, I kept getting run into by teenagers in stores and on sidewalks who couldn't both use their phone and navigate the world. That skill is now generally present and it's no longer a problem.

He didn't start getting it right until around episode 3, "The Reckless Newcomer".

> The vast majority of papers that I see with obvious LLM writing ticks are also frankly incomprehensible.

That's a classic problem with academic writing. There are too many papers where, after you finally figure out what they're trying to say, you realize they didn't have much to say. Relevant XKCD: [1]

What we seem to be getting from LLMs is more words per unit of useful knowledge. I wonder if academic papers are now bloated to allow for more ad slots. Many web sites with semi-academic content throw in an ad in the middle of the text every screen height or so. It's beneficial to the publisher if the author is wordy.

The same problem now plagues AI-generated videos, especially the pseudo-historical ones bloated to allow for more YouTube ad slots. These videos have a consistent structure. About 15% into the video, there will be something like "we have to go back to", followed by a large insert of filler material. Around the 50% mark, the relevant content starts. Around the 80% mark, the topic has been covered and there's some general blithering to conclude.

It's enough to make one long for the days of the neckbeard talking head with the big microphone.

[1] https://xkcd.com/451/

> This is something like handcrafted leather bag, quality assurance of parts like zips are okay but value lies in the craftsmanship rather than industrial production.

Nah. The value is in the branding.

There was an embarrassment a few years ago when it was discovered that one maker of fake Rolex watches was using the same movements as the branded version.

> One of my favorite New Criticism corollaries is that an author, after a work has been published, is not in a privileged position from which to speak on its meaning.

That's a line which was more appropriate in a era before the majority of novels had sequels. Now we read series where the plot extends across multiple books. Events in earlier books set up situations to be resolved in later books.

For a recent controversy in this area, ask Google "Is Dumbledore gay?"

It's partly a thickness problem. Sockets add height to boards.

Besides, since RAM isn't getting cheaper for a while, this isn't a big issue now.

> I opened this with an open mind, but it felt dead and AI immediately to me.

Episode 1 is. He's getting better at this. By episode 3, the characters have come alive and the scenery is really good.

Now go watch "Kono Express". Disney laid off an animator from the Star Wars factory, and he's cranking out stuff about as good as Disney Animation all by himself. With Unreal Engine and AI tools, of course.

It's been a long time since there was an ink and paint department.

Yes. But you have to go beyond page one of Google results because there is so much slop, human and otherwise. See [1] for a technical overview. See page 14 of [2] for the strategic and political aspects of who controlled what in a crisis. "Cheget No. 1 (the Soviet premier's portable terminal) had priority status: its owner could enact heightened level of alert and, after the early warning system registered the launch of U.S. nuclear weapons, give launch order (transmit codes unblocking the launch command)." That's the key point. The premier of the USSR did not have personal first strike authority.

Putin, however, has said that he now does. Trump has first-strike authority too.

The definitive book on this seems to be "Russian Strategic Nuclear Forces"[3], an English translation from MIT Press of a Soviet-era document. That's what the good sources cite.

[1] https://subhranil.com/the-dead-hand-system-a-technical-analy...

[2] https://npolicy.org/wp-content/uploads/2021/08/Controlling_S...

[3] https://search.worldcat.org/title/51969872

> The Russian’s dead man switch is the exact opposite. I wonder if it’s still in place.

Perimetr? That's misunderstood. The idea behind Perimetr was that the Premier of the USSR did not have first-strike authority. Only the Politburo did. After they got rid of Stalin, the later governments of the USSR were very wary of a leader with too much power. Further down, control was divided between the KGB, which controlled the warheads, and the Strategic Rocket Forces, which controlled the launchers.

The most the Premier could do on his own was enable Perimetr. If it detected actual nuclear explosions and could not communicate with any of the top command bunkers, it would, after a delay of hours, release the arming codes for ICBMs to lower levels of authority. So it only worked for a second strike where a first strike had taken out most of the leadership. That was more restrictive than the US system.

Today, both Putin and Trump have first strike authority. Sleep well.

That's an interesting point. Vending machines never became a big deal outside of Japan. Fast food was automated in 1964, with AMFare [1], but that was never deployed beyond one prototype.

Whether general purpose robots will change this is interesting. It looks like they're going to cost about as much as a car. (There are at least 18 makers of humanoid robots. The hardware situation looks good.) That's cheaper than humans. There are already companies that set up industrial robots and charge for them by the working hour, with all maintenance and setup included.

[1] https://www.youtube.com/watch?v=1Xop9py8zBY

Over on the entertainment side, video is getting really good. Yes, there's a huge amount of AI slop, but the good stuff is quite good.

- Kono Express. [1] Disney laid off an character animator from the Star Wars factory. The animator decided to see what was possible with the best modern tools. He's using both AI tools and Unreal Engine. There are six episodes now, and each one is a bit better than the last, as he gets a handle on the technology. The first episode looks like generic sci-fi stuff, at good quality but not yet perfect. By episode three, he has this nailed. It keeps getting better. Episode six [2] is a delight. He's now almost up to current Hollywood quality. One person, turning out half an hour of content every few weeks. Who needs Disney any more?

- Asian vertical dramas. 17,000 episodes per day. Apparently 55% of China's AI capacity is cranking these out. Search for "asian drama creation tools". Amusingly, Google and Bing have completely different tool lists. These are all software as a service products, with accounts and token charges. Currently, visuals are quite good, with some notable errors. Complex patterns on clothing don't stay consistent, for example. Continuity errors remain common. Automatically generated audio is currently terrible, at least in the English translations. Plots are so cliche that it's clear people are re-using the same prompts. "Heiress girl returns from countryside" is very popular. Here's a recent drama, about startup working hours.[3] The plot is silly but the visuals are OK.

- There's starting to be a thing for turning out movies based on old SF novels. So far, these seem unauthorized by the original author, but that will change. In time, every good book will have its movie.

The good stuff in this space is getting to be better than the Marvel Overexpanded Universe.

[1] https://www.youtube.com/watch?v=ikm5lP0zsi0

[2] https://www.youtube.com/watch?v=XklUIGp6jH0

[3] https://www.youtube.com/watch?v=J7b9xNEkhlc

The real job losses are probably in commercial art. If you're making an illustration for an ad, that's almost certainly automated now. Looking at when the decline started, it's about when Stable Diffusion got going.

Some of this stuff is pre-AI, Photoshop smart tools stuff. Especially clothing ads. Nobody has a live model pose with each item color any more. Or even each item. (I used to know actress/model/waitress types in LA who did that. Grind job: "OK, the blue sweater is next, and hurry it up, we have 50 to do before lunch". But it paid OK.)

> But internally we are using it for so many things, for example we have brought 3rd party analytics tools in-house (cost reduction and ownership over our data) by writing our own internal tools, we have setup custom systems to give ourselves better visibility over the business to try catch issues earlier, we use it to produce internal reports, etc.

How much of that is automation of "bullshit jobs"?

Yeah, Colossus and the Crab sucks.

We're not seeing full automation for things that ought to be automatable now. Customer service chatbots still aren't very good. Salesforce ought to be taking over entire marketing operations, not just supporting sales but replacing it. The medical office side of insurance ought to be automatic now. The pick and place side of warehouse operations should be robotic. Shelf restocking in stores ought to be robotic. All of those things have been demoed, but didn't go to production.

That's Air Hollywood's B-747 set.[1] They have other aircraft sets, an air terminal set, and supporting props, such as uniforms, beverage carts, and such.

[1] https://airhollywood.com/aviation-sets/wide-body-747/