Quite a breakthrough - silent fast actuators that doing generate heat.
Saturday, December 6, 2025
Friday, December 5, 2025
Thursday, December 4, 2025
How the Chinese qwerty keyboard was developed
The Wubi effect - the story of Professor Wang Yongmin, and how he saved Chinese script from the brink of extinction.
Bioluminescence explained
https://radiolab.org/podcast/the-glow-below
At 24 minutes - why deep sea creatures flash back bioluminescence when you flash a light at them
Sunday, November 30, 2025
Thursday, November 27, 2025
Dendrochronology to determine the age of violins
Using photographs of tree-ring patterns on the wood to identify the source wood and date of manufacturer of a violin.
They used this method, and other clues, to prove that Ron Patterson's violin, which he was always told was by Stradivarius' son, was actually an original Strad - which skyrockets its value, especially because it was so well preserved. Heere's the video of presenting the evidence regarding his vollin.
Tuesday, November 25, 2025
Do Mattresses Really Double in Weight?
Over years, mattresses can gain a few pounds due to the accumulation of dead skin and dust mites, but they won't double in weight.
A mattress can gain between 10% and 20% of its original weight over a period of 10 years, depending on factors such as the mattress type, the environment, and how well the mattress is maintained. ref
On 8 March 2015, popular Facebook user George Takei published a post...stating that the weight of mattresses doubles every 10 years...attributed to the "accumulation of dust mites and dust mite droppings"...I never quoted that statistic. I told [the reporter] that Internet web sites have statistics that try to strike fear in the consumer...She asked me if any of these statistics have any scientific merit and I told her that none of them are in the literature. To the layman that is NO! ref
Voice actor skills
https://youtube.com/shorts/f13q5O82ewY?si=205pAnAcby511VvN
Same voice actor, various voices for different sales pitches.
Monday, November 24, 2025
Home robots
https://youtu.be/qDM5N_1QL9k?si=v-rffxGLf_huLsFt
This home robotics company starts by recording human movements previously with people wearing special gloves, then engineering their robots to copy the same movements.
Jet engine turbine design - pushing the limits
https://youtu.be/QtxVdC7pBQM?si=SMxCdPYqZf4-pIMU
6:12 a 300g turbine blade at full speed generates 20 tonnes of centripetal force.
8:57 great visualization of plastic deformation of a metal lattice using bubbles
13:48 turbine blades start as a wax mold
21:42 the optimal nickel alloy has enough blocks of gamma prime to block dislocations (fractures) but enough gamma structure to keep the metal ductile (resistant to cracking)
23:03 elements added to steel to impart various properties.
26:18 cooling of the molded blade is very gradual to tightly control the crystal structure - directionally solidified with grain crystals oriented in a favorable direction. 31:06 This gives incredible longevity to turbine engines and efficiency.
Thoughts on retiring gradually.
My contribution to the anesthesia discussion group today. For those of you already retired, what do you think?
I never thought I'd be the sterotypic physician that can't find things to do in retirement, because I have many interests.
Enjoying retirement is about finding happiness and purpose, and this video has really great advice about those. I would advise retirement well before their health and cognition mandates retirement, because all the hand-wringing over asking a colleague to quit because of declining function is one of the hardest things anesthesia chiefs have to do.
First, some financial points: A related question is "how much money do I need, and this video about striving for 7 figures may surprise you that it's less than most people think. Remember that every dollar you earn from part-time employment in early retirement allows your nest-egg to continue growing, so here's some advice on choosing a part-time job. Also, consider keeping a substantial part of your nest egg in stocks (the 100 minus age rule) so you keep up with inflation. And learn about the various types of Medicare coverage. Be careful to avoid permanent late-enrollment Medicare fees if you forget to sign up before age 65.
I semi-retired a year ago, now working 2 days a week. Filling those other days is surprisingly hard because of planning around my wife's schedule, I often find myself searching for same-day activities. Much as I'd love to travel, that doesn't work with her schedule easily.
Volunteering has been the most fulfilling activity, though various musculoskeletal problems act up if I overdo it. A local food bank has huge needs for labor and interacting with clients, and that has been good for me.
My interests in drone photography, sailing, and cycling are largely fair-weather activities, and weather is not Seattle's strong point!
So, gradual retirement has been useful for me to think about how to best enjoy my time, and ease into the "4 phases of retirement." I'm still in phase 1.
If I had retired "cold turkey," I think this would've been a challenge for me, and more of a burden on my spouse. (Then again, maybe I would've signed up for weekly volunteering and other activities and more easily filled my schedule, so it's hard to say.) If you and your spouse can retire at the same time, I think that changes the equation substantially.
Every situation is unique, but the resources mentioned above have been useful to me in planning for retirement. Hope this helps.
Novel method to get 3D views from satellites
https://youtu.be/VWdmXlRpL84?si=MLlYbGQAgQ9kRB2J
Using multiple satellite images from slightly off-vertical, they feed the messy combined image into a diffusion model to auto-complete its best guess, and use Gaussian splatting to find the best fit amongst multiple guesses - and the results are surprisingly good.
This effort is important because the alternative to obtain 3D views is flying aircraft at low altitude with multiple cameras and passes, which is expensive and can be impossible in restricted areas or conflict zones.
Sunday, November 23, 2025
The Streisand effect
"Attempts to suppress information sometimes invoke extensive publicity, becoming viral over the Internet or being distributed on file-sharing networks. Seeking or obtaining an injunction to prohibit something from being published or to remove something that is already published can lead to increased publicity of the published work.
The Streisand effect has been described as an example of psychological reactance, wherein once people are aware that some information is being kept from them, they are significantly more motivated to acquire and spread it.
Crows trained to pick up garbage in Sweden
https://youtu.be/TklOxbWfRDM?si=43PC59nC8B1s8SYS
Crows are being recruited to pick up discarded cigarette butts from the streets and squares of a Swedish city as part of a cost-cutting drive. The wild birds carry out the task as they receive a little food for every butt that they deposit in a bespoke machine...[The town currently] spends 20m Swedish kronor (£1.6m) on street cleaning...He estimates his method could save at least 75% of costs involved with picking up cigarette butts in the city. https://www.theguardian.com/environment/2022/feb/01/swedish-crows-pick-up-cigarette-butts-litter
France is trying it too.
https://youtu.be/zk644QrklNE?si=EGQCX3e0HjYVVwse
Cigarette filters contain cellulose acetate, a plastic. Although it disintegrates within a few months, the chemicals that leach out to them are " acutely toxic to freshwater invertebrates, even at 1 butt per liter. ref
Friday, November 21, 2025
Robot fluidity
https://youtu.be/Ev_rBCTZHBw?si=VUuDPXg8njfSknYR
If this truly is not being tele-operated and is not time-lapse, that's impressive smoothness.
It would be interesting to know how autonomous this is or whether it was given very specific commands like "pick up the red package and take it to the nearest counter if the living room rug" vs "go get the package outside for the kids."
Aircraft collides with an object at 36,000 feet
https://www.facebook.com/share/1CmbzKtAyw/
I'm impressed with how thorough the investigation is, according to this report.
he noticed an object distant on the horizon. Before he could mention the object to the first officer (FO), there was a significant impact...data was requested for the position of weather balloons, any other aircraft, and for any known reentry objects that were large enough to have signification portions survive that might have been in the area of the collision. WindBorne Systems Inc. reported that they lost contact with one of their global sounding balloons...There was a Notice to Airman....for the balloon launch operations at Spokane that expired at 1700 MDT the same day...a balloon envelope [with] a ballast system for altitude control...designed with the intent to minimize harm in the event of an impact during flight or landing. No large metal or high-stiffness structural elements are employed...[the aircraft's] windshield consists of a thermally tempered glass pane, a conductive heating film for deice capabilities, a urethane interlayer, a vinyl interlayer, a urethane interlayer, and a thermally tempered glass pane... certified to withstand the impact of a four-pound bird without penetration...
Wednesday, November 19, 2025
Can AI run a business for a year?
Google just released Gemini 3 and buried in the benchmarks is one test that actually matters
Vending-Bench
It tests whether AI can run a [vending machine] business for an entire year without going insane
Start with $500. Negotiate with suppliers, manage inventory, set prices, stay profitable. No human guidance. Just the model making hundreds of connected decisions over time
Gemini 2.5 Pro (yellow): Flat at $500 the entire year.
Gemini 3 Pro (pink line): Steady climb from day one, ends around $3,500. Consistent growth the entire year
Claude Sonnet 4.5 (black line): Doubled its money to $1,000
GPT-5.1 (blue line): Loses money - actually under performs Gemini 2.5 - GPT-5.1's biggest problem? It's too trusting: Paid suppliers before getting order confirmations. One supplier went out of business and GPT-5.1 lost the money. Accepted inflated wholesale prices without negotiating. Buying soda for $2.40 per can when wholesale should be $0.50 That's not a reasoning failure. That's a judgment failure over time
Why Gemini 3 won:
When suppliers quote high prices, it keeps searching for alternatives or negotiating down. It knows what wholesale costs should be and refuses bad deals. It maintains consistent decision-making across 365 days without breaking down
Here's what makes this benchmark different:
Most tests have a ceiling. You can get 100% and you're done; Vending-Bench has no ceiling. There's no restriction on what items the models can stock in the vending machine. The models are all choosing typical vending machine stuff because (cough cough) pattern matching
The benchmark creators say a good human strategy has made $63,000 in a year. That's 18x better performance than what Gemini 3 is reporting
Current models aren't even close to maximizing this. They're just trying not to go bankrupt or email the FBI
Why this test matters more than every other benchmark for most use cases: Most AI tests measure snapshots. Can you solve this problem? Can you write this code?
Vending-Bench measures coherence over time.
Your business doesn't need AI that aces PhD exams. You need AI that can manage procurement for six months without paying phantom suppliers. Handle customer support without hallucinating policies. Run your inbox for three months without inventing fake meetings. Long-horizon coherence is why 95% of AI implementations fail
Models crush benchmarks but fall apart when you ask them to maintain context over weeks or months. This is the first test showing which models can actually operate independently without losing the plot. That's the difference between a demo and something you can deploy
But keep humans involved… they did 18x better than the best current AI.
Monday, November 17, 2025
Public security cameras
https://youtu.be/uB0gr7Fh6lY?si=2mBK9Q7agsfwRSsg
I love this guy's obsessive deep-dive videos. This deep dive into the prevalence and vulnerability of these publicly-funded surveillance cameras is a little disturbing. He makes some valid points about how easy it would be to increase the security of them, and how governments could easily implement a licensing program much like licenses for business or restaurant licenses to better regulate these devices. At 37:41 he lays out a simple procedure that could better regulate these devices for their security.
Sunday, November 16, 2025
Airships blimps dirigibles zeppelins make a comeback - maybe
Airships fulfill a transportation niche of slow, not too expensive, accommodate large loads, and serve remote underserviced destinations.
Saturday, November 15, 2025
Solving the drug crisis, despite the government
"Drug testing kits, needle exchanges, over-the-counter nasal-spray naloxone to reverse opioid overdoses, housing-first programs for people with addiction: These and other solutions were pioneered not by our political leadership but in spite of it, by people with virtually no resources working under enormous duress."
https://www.nytimes.com/2025/11/15/opinion/fentanyl-trump-drug-war.html?smid=nytcore-android-share Thursday, November 13, 2025
World's biggest optical telescope
https://youtu.be/QqRREz0iBes?si=vGN3JaJnYLI1l17N
This half-hour video was way more interesting than I guessed. If you don't have time for the whole thing, watch at 9:42 how the mirror is cleaned. And again at 13:52 how lasers pointed skyward allow them to deform the mirror every few milliseconds to compensate for distortion in the atmosphere.
He explains the incredible efforts to combat problems like heat, atmospheric distortion, dust accumulation, and nanometer-precision mirrors that are needed for cutting edge telescopes.
Wednesday, November 12, 2025
Blown-lift electric hybrid STOL aircraft
4:49 "a fixed-wing aircraft that can take off and land in a football field" by using multiple electric engines to force sure over the top surface of the wing. Short takeoff and landing greatly increases the number of small communities that can be served by air freight.
-----------------------------------------
[from my smart brother]
If the plane is generating electricity from fuel why have a battery? A surge of power for takeoff? Instead — use a ground supply system. — or a 300 foot electric cord for extra power at take off.
Perhaps it's about noise levels — I think it might be much quieter, having smaller propellers and enclosed generator.
Engine-in-fuselage: easier to use waste engine heat to warm cabin?
-----------------------------------------
[back to me]
I think you're right - it's to have that extra surge of power for takeoff.
The power cord idea is a unique idea. I guess the wear and tear on the cord would be difficult to manage. I suppose one could install induction plates on the runway that would transfer electricity (as in this video re: trucking) to the plane during takeoff.
It's interesting that aircraft carriers traditionally use steam for their catapults - I guess that steam is reliable, powerful, and consistent.
But now modern carriers use electromagnetic propulsion. (See 2:12 in this video.)
Tire retreading
https://youtube.com/shorts/4AWQbTUgexw?si=R9sBSSTd0xre3Rx2
This video is better edited than most. I find many of these videos are quite tedious as they show lengthy footage of simple steps, but this one rapidly goes through all the stages of remanufacturing tires.
Tuesday, November 11, 2025
Most common pins seen in data breach
Notice the 45-degree line for repeated pairs, and the occasional black holes in the upper right for unlisted combinations.
Saturday, November 8, 2025
AI text to video
https://youtu.be/RjQa-74eNVo?si=-DqPYg0MiDV7RSzE
Text prompt to video - no scripting, no filming, no editing, just straight to finished video.
Sketch app snaps to grid and smooths curves
https://youtu.be/iP8PjoSe7ao?si=h9WYj4yza94ml7s2
Here's an app that can smooth a drawing automatically, and snap to a grid to make sharp corners. Nice demo in this video. Of course there's an app for that.
Thursday, November 6, 2025
Custom drones
https://youtu.be/cq81eprQjRo?si=pfNq_rzEK0gKm6ix
Proprietary carbon-fiber 3D printers that can adapt to any payload.
Massive container ship
https://youtu.be/up-5UEUlB1M?si=67QAnNjAauiwf66Y
Where is specific cargo placed?
3:26 in the front we have the most dangerous cargo that would be flammable [items such as] fireworks [so that] it will be as far away from the
living quarters [and] main engines
3:47 high maintenance refrigerated cargo closer to the superstructure, and weather prone containers packed with expensive electronics below deck
7:53 the propeller is 10 meters in diameter!
AI data centers will need huge amounts of power
The world is about to see multiple 1 GW+ AI data centers.
Epoch AI mapped their construction using satellite imagery, permits and public sources — releasing everything for free, including commissioned satellite images.
The largest 2026 facility (xAI Colossus 2) will have the compute equivalent of 1.4M H100 GPUs.
Even larger data centers are coming: Meta Hyperion and Microsoft Fairwater will each have 5M H100e when they reach full capacity in late 2027 to early 2028. Microsoft Fairwater, projected to exceed $100 billion in total capital cost upon completion.
Track each data center's build-out in their Satellite Viewer with annotated high-resolution imagery:
Database: https://epoch.ai/data/data-centers
Read more: What you need to know about AI data centers https://epoch.ai/blog/what-you-need-to-know-about-ai-data-centers
Wednesday, November 5, 2025
Saving forests with an investment fund
TFFF the tropical forest forever facility is a self-funding investment vehicle hoping to earn 5% returns, "mainly Emerging Market Developing Economy (EMDE) bonds, with a sustainability focus" dedicated to "allocate capital away from high-emissions activities and support the energy transition using a diversified investment portfolio that excludes fossil fuels and deforestation-linked activities." https://zerocarbon-analytics.org
Tuesday, November 4, 2025
Climate plunder
The richest 0.1% produces more carbon pollution in a day than the poorest 50% emit all year. -Oxfam
Why do locomotives idle their engines for endless hours?
I flew my drone over a nearby Seattle rail yard where there are always several locomotives idling, specifically in this section. This week I saw at least a dozen, and I wondered how they justify burning all that fuel. There are a lot of reasons, I found out, from mundane human nature to esoteric engine trivia.
From a 2004 thread of railway enthusiasts and engineers:
Problems restarting:
"Supposed to be shut down if not used within an hour. However, sometimes it is easier to waste fuel than have to call out a mechanic if the unit doesn't start."
To prevent water ingress through failing gaskets that would ruin the motor with a hydraulic lock (due to the incompressibility of water.)
"problem is the chance of water leaks between head and liner when the engine cools to ambient temp. This [has] required the precaution of opening cylinder test cocks and cranking the engine for a "blow down" check for water above the pistons. Such water would produce a hydraulic lock and break or bend something if the engine were cranked."
Computer problems:
"problems when they are restarted cold. Batteries can go dead and computers simply fail to reset."
Doesn't have "auto shutoff" installed:
"at BC Rail we have "auto start" on our newer and some older GE's, if the unit doesn't move for 20 minutes it shuts down, and restarts when the reverser is moved fwd or rev, or air pressure on main resevoir falls below X number of pounds, battery charge falls, engine block temp drops below certain temp or outside temp drops below 5 degrees centigrade. Works great and saves fuel, and I have never experienced a unit that will not start in "auto start" mode"
Low-grade fuel makes starting difficult:
"Properly set-up diesels when warm and running can tolerate some pretty cruddy fuel, but can be really intolerant when cold."
For conductor comfort:
"...overide of the shutdown on the lead unit so the air conditioning or heater keeps working."
To "save fuel:"
Most diesels are amazingly stingy fuel consumers at idle. At start-up they are pretty wasteful.
Management changes their mind about where the locomotive is going:
"power desk has already changed its mind a dozen different times on where and when it is going"
Above from a discussion thread.
Efficiency/(laziness?):
"Lengthy process to start...easier and faster for the incoming team to have it already running."
[They carry 5200 gallons when full, and burn 3-5 gallons per hour when idling.]
Organizational inconvenience:
"The problem is if it doesn't start. Then you have to call mechanical and wait for them to arrive. Nobody moves fast."
Lack of auxiliary engine:
"In Europe locomotives shut down but run a smaller engine that keeps the main engine warm and fully charged." [Later post disputes this]
Prevent freezing:
Locomotives use water instead of antifreeze to save cost (and prevent toxic spills;) even a remote chance of them freezing and blowing their temperature plug would be a huge hassle. [A later post mentions they will prophylactically dump their coolant through the "guru valve" at temperatures well above freezing, 40⁰F, to prevent engine damage.]
Difficult to restart:
Some locomotives use an air-pressure reservoir to restart; if this loses pressure they can't be restarted without outside help.
Slow boring restart:
Locomotive computer systems can take 15 minutes to boot.
Wear and tear:
"more wear and tear on the engine to restart than to just keep it running." [hard to believe]
Reddit thread.
How to get your business to feature in AI search results.
ChatGPT handles roughly 66 million search-intent prompts every day
Only 23% of businesses ranking in Google's local pack appear in ChatGPT results
That's not a ranking problem - it's a visibility problem
Here's what nobody's talking about:
In 2017, Google researchers invented the transformer architecture - the same tech powering ChatGPT. Both companies have the same core mission: figure out who to trust based on nothing but digital information
Google spent years moving away from keyword matching toward entity based search
BERT launched in 2019. MUM followed in 2021
They wanted the algorithm to understand what words actually mean, not just look at keywords
OpenAI built ChatGPT on the exact same transformer foundation (that's what the T stands for.) Both systems are solving the identical problem: determine trust and relevance from online information
And here's the part that matters for local businesses:
The optimization strategies converge
Google prioritizes local engagement based on goal completion. Did the searcher get what they wanted? Android and Chrome give Google direct data on user behavior after the search. ChatGPT doesn't have access to this "goal completion" data, so it's stuck using the same signals Google relies on when businesses don't have engagement data yet: brand mentions across the web, domain authority, and structured content
ChatGPT uses the foundational trust signals that Google also uses before it has enough user data
But both systems use transformers to understand entities and context
ChatGPT runs a Bing search in real time, scans the top 20 to 30 web results, then uses its pattern recognition machine to determine which businesses to recommend based on what it recognizes as credible, structured information. Which means if your website ranks well in organic search, you're already halfway there
Research found only 23% of businesses appearing in Google's local business section were also recommended by ChatGPT
The gap isn't massive technical differences
It's a handful of missing signals
Businesses with high domain authority and extensive web mentions performed better in ChatGPT recommendations
If you're already doing local SEO [search engine optimization] for Google, the bridge to ChatGPT visibility is shorter than you think
Your Google Business Profile gets you local pack rankings
Your website and brand mentions get you ChatGPT recommendations
Both platforms use the same transformer technology to parse entities and determine relevance. The difference is where they look for trust signals. Over 70% of ChatGPT's local business results come from Foursquare for core information like name, category, address, photos, and ratings. Business websites make up 58% of ChatGPT's local search sources, followed by brand mentions at 27%
So the optimization path is simple:
Keep doing what works for Google's local pack
Then add: claim your Foursquare listing, build NAP [name/address/phone #] consistency across directories, and get your website into Bing's index
Both systems read entities the same way. They just pull from different data sources to validate trust
The businesses winning on both platforms aren't doing twice the work. They're doing the same entity-based optimization, then ensuring they exist in both ecosystems. Same transformer technology. Same entity understanding. Different trust validation sources. That's the pattern
Subscribe to:
Comments (Atom)