Dave Kammeyer

@DrPhiltill Yeah it's funny how everyone is different in this. They gave us a pulse ox, but didn't tell us when we should turn on the oxygen. I was down to 58 when I finally turned on the O2, and I felt coherent then, but I later found out that that was a pretty dangerous situation!

Yeah, we did a worksheet, too. The guy next to me had the most massive grin on his face that I’ve ever seen, but wouldn’t respond to any commands to put his mask back on. Totally lost useful consciousness. The instructor finally held the mask on his face and turned up the O2 flow until he became responsive again. Good times 😄

I got to do this high altitude training at the Sonny Carter Test Facility in Houston. You take off your mask and record your symptoms along the path to losing “useful consciousness”. My first two symptoms were tunnel vision and fumbly hands. One guy in our group went fully incoherent.

@ramez @EnergyJvd Always shocking to me how long people will fight a losing battle trying to improve their terms of surrender. WW2, Civil war...

@dccommonsense Super important life habit. We spend almost all of our time looking at screens or otherwise receiving input that dictates our thoughts. It's easy to get a lot of unresolved crap rattling around your brain when you never just let it go where it needs to go.

I'm not much of an advice giver,but if you want to develop an ability that will give you an advantage in the future develop your contemplation skills. Learn to be alone and just THINK. Be bored. Make your brain do things it need not do when you feed it stimulation constantly.

@aakashgupta Right but the water is coming from the Mississippi River. That billion gallons the datacenter uses in a year goes into the Gulf every 3 minutes. And by the way the steam is recycled. It goes into the air and falls as rain in the eastern U.S.

Google’s single data center in Council Bluffs, Iowa consumed 1 billion gallons of fresh water in 2024. One facility. One year. Enough to supply every home in Iowa for five days. The reason they need fresh water is pure chemistry. Evaporative cooling towers work by running water over hot surfaces and letting it evaporate. 80% of the water a data center pulls in literally vanishes into the atmosphere as steam. You can’t recycle steam. The remaining 20% becomes concentrated mineral waste. Calcium, magnesium, silica. Every cycle through the cooling loop makes the water more corrosive. After enough passes, it starts clogging pumps and eating through heat exchangers. Multi-million dollar equipment destroyed by limescale. Recycled wastewater carries even more of these minerals from the start. You could treat it, but less than 1% of U.S. water is recycled. Most cities don’t even have separate pipes to deliver reclaimed water to industrial customers. A data center wanting to use recycled water would essentially need to build its own treatment plant on site. Meanwhile, municipal potable water costs almost nothing. So they just drink from the tap. Across all its data centers, Google used 8.1 billion gallons in 2024, nearly double what it used three years earlier. The company claims its water stewardship projects “replenished” 4.5 billion gallons. Those projects aren’t even in the same watersheds where they’re pulling the water. Same playbook as carbon offsets. Consume locally, offset globally, call it sustainable. The trajectory is wild, to say the least. U.S. data center water consumption could quadruple by 2028. That’s 68 billion gallons for cooling alone, before the 211 billion gallons consumed indirectly through electricity generation. Two-thirds of new data centers since 2022 are being built in regions already facing water scarcity. Nobody’s asking why they use fresh water. They’re asking what happens to the towns sharing a water main with a facility that drinks like 50,000 people showed up overnight.

@piyushahuja_in @yaroslavvb No the RL is a step beyond just predicting the next token on the web. You could still view the resulting transformer as a next token predictor, but it would be for human-judged "good responses" instead of the next token on the average web page.

@davekammeyer @yaroslavvb But isn't human reasoning not exhibited nless the odel's are RLed in post training? is RL also part of search for efficient Shannon compression?

The biggest surprise of the last decade: in our search for efficient Shannon compression (figuring out the code-book for web data) we accidentally discovered the algorithm of human reasoning

@piyushahuja_in @yaroslavvb The algorithm is the model. The algorithm is found through gradient descent, but the interesting thing is that compression is all about predicting likely sequences. That's exactly what a transformer does, so compression=prediction=intelligence.

@yaroslavvb Would you be willing to elaborate? Is gradient descent 'the algorithm of human reasoning' herein?

@Itsjoeco Except for the massive amount of weapons grade Uranium in all of them.

My proposal would be to have the US Navy begin commissioning and operating nuclear reactors on land INCONUS. We could get 10 GW nuclear from 20 Ford-Class carriers. Yeah, it would be weird to have naval officers stationed in Dubuque but energy is approaching a national emergency and we are the only people currently building and safely operating new nuclear power at scale. So just do it for a little while.

@wolfejosh Similar, in 2023, but how did you buy it? There seemed to be rules against investing in Korean stocks. It appeared I'd have to open a Korean brokerage account which didn't seem easy/impossible for a non-resident.

From fall 2024 to today….when I privately pitched SK Hynix

@DJSnM A combination of MLAT and the reported position would reduce the spoofing potential a lot. If MLAT disagrees with reported position, report the MLAT and the discrepancy. If only one receiver sees the transmission and other stations should have, flag it.

ADS-B is public and unencrypted, anyone can generate radio transmissions that will be picked up by amateur network and reported to aggregators. ATC still use traditional Mode-C/S transponders that get pinged by ground stations making position harder to spoof.

@0xmjt @nntaleb The 20 MET hours are something like 3 clock hours. My guess is that the stronger U shape of cycling is because a little cycling helps your body, but the more you do it the more risk you have of being hit by a car. Would be interesting to see what a stationary bike looks like.

@nntaleb those cycling for more than 20 hrs per week very likely have something else wrong with their lives…. 😀

"J" (or "U") curves are common in medicine. But when you see in a paper a "W" ("a little bit helps, some more harms, but a lot more helps") w/a mention of both "p" &"nonlearity", you must question it--particularly when it contradicts other well powered studies on dose response.

@nntaleb You can fit a u shape inside their error bars though.

@patrickc Worst part for Google Maps is that they demand that you participate. They pop up a box that says "Police still there? Yes/No", with no X button, and the box covers up your ETA until you push one of them. Deliberately distracting drivers so others can avoid speeding tickets.

The "police ahead" feature in mapping applications is peculiar. What exactly is the point here? Do we want road safety enforcement to be better or worse? Should these applications also provide notifications for ticket inspectors on public transit?

@JensWalterEU @RealJakeBroe It's still there. One of the runways was decommissioned, but it's now the Narsarsuaq airport. I think it still gets some military traffic. If you've ever read Fate is the Hunter (you should), it's the famous Bluie-West-1

@RealJakeBroe Found this on Flickr, scanned images, taken during WWII. Now abandoned since 1958: Narsarsuaq Air Base, Photographer Clyde E. Gray: flic.kr/p/2pBh7yL

During the Cold War the United States military had 15,000 soldiers stationed in Greenland. That presence has been reduced by 99% to just 150 soldiers today. Denmark has given the US exclusive and full military access to Greenland but the US does not use it. This has nothing to do with security.

@BenniKim Where does this chart come from?

Frequency matters a lot. A F-35 is not very stealthy to radar systems operating in VHF band. But it is very very good against higher frequency to the point of being almost invisible. So the shitty 1970 soviet-era VHF radar actually might get a change to see it. This was documented in this australian military magazine. ausairpower.net/SP/DT-Rus-VHF-…

The "stealth" (Radar Cross Section) advantage of the F-35 vs other fighters is nowhere near as much as you think it is in operational conditions, especially against a multistatic radar installation using frequencies below 1 Ghz. F-35 with 2x wingtip AIM-9X F-16 with 2x wingtip AIM-9X Rafale with 4x Meteors The moment missiles are carried outside and/or the radar illumination comes from a 25°+ angle on the side, the stealth is gone. And given the classic radar equation, 6 dB better than an F-16 or a Rafale doesn't buy you a lot of time.

@mbw61567742 I've been giving them piecemeal around the dinner table and in the car for the past couple of years 😀it is great to have the slides though.

@davekammeyer I didn't video any of them. This is your chance to give the lectures for your kids! Maybe video them? p.s. I'll probably soon be posting links to my wife's extremely successful intro stats courses. These will be videos.

I taught a course at the intersection of physics, philosophy, and history. I think it was pretty good. The blog linked below gives a very brief description and links to slides for the first half.

@mbw61567742 It was really, really great! It looks like you've added to it some since I took it ~27 years ago... You didn't happen to video the lectures did you? My kids are getting to the age where they are interested in this stuff and I haven't seen anything nearly as good online.

@davekammeyer Thanks! It's hard to describe how much that means to me.

@ebarenholtz @markopolojarvi @niko_kukushkin If you had a big enough corpus of genomes, I think the model would eventually infer the existence of the cellular machinery, chemistry, ecosystems, etc. I suspect that high perplexity. in DNA models is mostly because there are so many synonyms in the DNA code.

This actually clarifies things very nicely. You could train a transformer on the entire corpus of nucleotide sequences ( people have) and it would never be able to predict the next base with high accuracy in an open-ended way. The reason is that the determining factors are not fully in the sequence. DNA continuation depends on external biochemical machinery and environmental conditions. Language is different. The information needed for continuation is fully internal to the linguistic structure itself. The “rules” of generation are embedded in the corpus, not executed by an external system. That’s precisely what makes language unique among generative systems: its continuation arises entirely from its own topology.

People still don’t seem to grasp how insane the structure of language revealed by LLMs really is. All structured sequences fall into one of three categories: 1.Those generated by external rules (like chess, Go, or Fibonacci). 2.Those generated by external processes (like DNA replication, weather systems, or the stock market). 3.Those that are self-contained, whose only rule is to continue according to their own structure. Language is the only known example of the third kind that does anything. In fact, it does everything. Train a model only to predict the next word, and you get the full expressive range of human speech: reasoning, dialogue, humor. There are no rules to learn outside the structure of the corpus itself. Language’s generative law is fully “immanent”—its cause and continuation are one and the same. To learn language is simply to be able to continue it; the rule of language is its own continuation. From this we can conclude three things: 1)You don’t need an innate or any external grammar or world model; the corpus already contains its own generative structure. Chomsky was wrong. 2) Language is the only self-contained system that produces coherent, functional output. 3) This forces the conclusion that humans generate language the same way. To suggest there’s an external rule system that LLMs just happen to duplicate perfectly is absurd; the simplest and only coherent explanation is that the generative structure they capture is the structure of human language itself. LLMs didn’t just learn patterns. They revealed what language has always been: an immanent generative system, singular among all possible ones, and powerful enough to align minds and build civilization. Wtf.

@levelsio This was pretty much the Sonder model. I stayed there a couple of times and it was good for the price, but they just went broke…

I'd just create an entirely new hotel chain No front desk, no lobby Payment and booking via app only Room keys via Apple / Android Wallet Hourly rates Coworking in building Powerful AC in every room and space Only cleaning staff