Tag: science

Sleep Enhances Motor Memory Consolidation in Piano Practice and Beyond | January 25 2025, 18:10

I have long noticed an interesting feature. You practice a piece on the piano and when progress seems to stall, you quit, but the next day when you sit down at the keyboard, suddenly, the difference from yesterday is like night and day, even though all you did was sleep.

I found a study (Journal of Neuroscience) suggesting that the proximity of sleep to practice plays a significant role. The researchers hypothesized that motor memories – the brain’s way of preserving skills and actions – not only consolidate over time, but can significantly improve if sleep follows soon after practice.

This hypothesis was tested through a series of experiments involving 290 right-handed individuals. Participants had to move a cursor on a computer screen to hit targets using a joystick. The complexity arose because sometimes the movement of the cursor was altered by optical rotation, which forced participants to adjust their hand movements to accurately hit the target. Different groups of subjects underwent this at different times, and those who did it before sleep showed the best progress.

It turns out that engaging in activities involving a significant portion of motor memories—like playing the piano or participating in complex-coordinated sports—right before sleep is beneficial. You could either adjust your sleep or the timing of these activities.

Navigating the Nexus: Harari on Information Networks and AI | January 19 2025, 16:42

Well, I finally finished reading Yuval Harari’s “Nexus: A Brief History of Information Networks from the Stone Age to AI.” Quite interesting, although I was expecting more controversy, especially from a historian who likely doesn’t quite understand what’s under AI’s hood. But it turned out to be quite good. The theses extracted from the book without context sound quite odd. For example, Harari writes that it would be more accurate to call AI Alien Intelligence—not because it possesses consciousness, but because its methods of thinking and problem-solving are completely alien to our human experience.

It’s particularly fascinating how he views political systems as information structures: democracy with its decentralized flow of information and autocracy with its drive for control. An attempt to understand how information governs us, our choices, and our societies.

Overall, the parallels drawn between the AI revolution and past historical events are quite intriguing. Harari cites examples of how social network algorithms aimed at increasing engagement have contributed (and continue to contribute) to the spread of misinformation and the fomenting of hatred. He references real cases, such as the tragic events in Myanmar, where Facebook was used to spread falsehoods that led to violence. Because falsehoods positively affected engagement—they simply got more clicks—and this function was optimized by Facebook for better advertising metrics. Ultimately, a relatively simple AI behind the newsfeed led to bloodshed.

I liked Harari’s example from Nick Bostrom’s book “Superintelligence,” which is a thought experiment. Bostrom suggests imagining a paperclip factory acquiring a superintelligent AI, and the factory manager assigns the AI, seemingly simple, the task of producing as many paperclips as possible. In pursuit of this goal, the computer takes over the Earth, destroys all humans, sends expeditions to capture additional planets, and uses vast resources to fill the entire galaxy with paperclip factories. The essence of the thought experiment is that the computer did exactly what it was instructed to do.

Of course, it’s hard not to notice that Harari sometimes succumbs too much to alarmism. For example, he describes AI as something already spiraling out of control, though there are many technical constraints that inhibit its development. Nevertheless, his book is not a prediction, but an invitation to a discussion. A discussion about how technologies change our understanding of truth, freedom, and even reality.

P.S. I also liked this fragment:

“…It’s no wonder that politicians around the world spend a lot of time and effort recounting historical narratives. The aforementioned example with Vladimir Putin is hardly an exception. In 2005, UN Secretary-General Kofi Annan first met with General Than Shwe, the then-dictator of Myanmar. Annan was advised to start the conversation first to prevent the general from monopolizing the talk, as the meeting was supposed to last only twenty minutes. But Than Shwe spoke first and almost for an hour about the history of Myanmar, leaving almost no opportunity for the UN Secretary-General to speak.”

Surprisingly, Harari did not include interviews with Tucker Carlson 🙂

The Unpredictable Rise of a Small Jewish Sect Over the Roman Empire | January 19 2025, 14:54

“How many Romans or Jews in the time of Tiberius could have predicted that a small Jewish sect would ultimately conquer the Roman empire, and emperors would forsake the old Roman gods to worship a crucified Jewish rabbi?”

Indeed, a good question. As far as I am aware, at present, there is no religion that continues the traditions of ancient Roman or ancient Greek polytheism in their original form. Curiously, why is that?

I think that religions without centralized power simply stand no chance. On the other hand, what about Hinduism and Taoism? I’m not well-versed in this subject, but it’s interesting. I had never contemplated how it turned out that a dominant religion across a vast territory was completely obliterated.

Exploring M-DISC for Long-Term Photo and Video Archiving | January 19 2025, 13:05

So you have a photo and video archive and want to preserve it for your grandchildren and great-grandchildren. Where to store it? I did some research and found only one option, which is rather affordable overall.

First, why are the others not as good? Any cloud solution is potentially unreliable, as it depends on the long-term stability of the provider, risk of data loss due to cyber attacks or infrastructure failure, political processes, good air defense, and the need for ongoing payments.

CD/HDD/SDD/tape are generally better when maintaining control over humidity/temperature/lighting, but still, no one guarantees more than 20 years. A flash drive could start losing data as soon as in five years. Magnetic tapes are better, but they also gradually demagnetize, and accessing data is relatively slow — a special reader, which is quite niche, is necessary, and it’s unclear what the situation will be in 40-50 years, and they won’t last much longer than that.

It seems that the only remaining option is M-DISC discs. It is claimed that M-DISC has a lifespan of 1000 years. It is made from a durable inorganic material similar to stone, which does not degrade over time, heat, or light. It is supported by many modern DVD and Blu-ray drives, although an M-DISC-compatible drive is needed for recording. M-DISC capacity is 4.7 GB (DVD-R), 25 GB (BD-R), 50 GB (dual-layer BD-R), or 100 GB (BD-XL). Clearly, these are all write-once, essentially like carving into granite.

As for the cost — an external USB drive with BDXL M-DISC recording capability costs about $100, although there are options for $50, and BDXL discs cost about $10-12 each (100 GB).

Reflections on 2024: AI’s Ascent and Google’s Evolution | January 18 2025, 23:18

Reading an article in Wired, the link will be in the comments. It’s fascinating to see how people from the past envisioned 2024.

In 2002, Kevin Kelly, one of the founders of Wired, attended a small party organized by Google, which at that time had not yet gone public and was solely engaged in search technologies. At this event, he struck up a conversation with Larry Page, Google’s co-founder, who later, in 2011, became the CEO of the company (and left the position in 2015).

Kelly asked Page a question that seemed sensible at the time:

— Larry, I still don’t understand. There are so many companies doing search. Free web search? Where is this going?

At that time, Google had not yet implemented its advertising sales scheme through auctions and had not made major acquisitions like YouTube. Many, including Kelly, doubted that Google would last long.

Page then replied:

— Oh, actually, we are making AI!

Years later, when Google had acquired 14 companies working in the fields of AI and robotics, Kelly often reflected back on that conversation. At first glance, it might seem that Google uses its AI resources to improve search, because search then generated 80% of the company’s revenue. However, Kelly wrote in his 2014 article that the opposite is true: Google uses search to enhance its AI.

Every time a user enters a query, clicks on a link or creates a link on the internet, they train Google’s AI. For example, by searching for “Easter bunny images and selecting the most appropriate one, a person teaches the system what an Easter bunny looks like. As of 10 years ago, when the article was written, 1.2 billion users made 12.1 billion queries daily, continuously training AI.

Interestingly, in that 2914 article Kelly was confident that by 2024, Google would create AI and it would gradually become the main product.

Well, here we are in 2024, and as of today 82% still comes from the same advertising. But still, 2024 became the year of AI and Google is among the key players (along with a dozen others).

Interestingly, the Larry-predicted formula from the article – add AI to everything indiscriminately and you get a new startup – worked out

“…In fact, the business plans of the next 10,000 startups are easy to forecast: Take X and add AI…”

He writes in 2024 that for AI, progress in three areas is critical:

1) GPU/parallel computation for training models

2) Big Data for training models

3) Algorithms that we have yet to conceive

“… This perfect storm of parallel computation, bigger data, and deeper algorithms generated the 60-years-in-the-making overnight success of AI.”

Exploring the Global Reach and Unique Development of SQLite | January 18 2025, 17:00

Discovered a plethora of interesting facts about SQLite. For those unaware — it is likely the most widespread database in the world. It’s utilized everywhere: from smartphones to cars and rockets. The scale of deployment — over _a trillion_ instances, billions of active copies worldwide. In your phone, dozens of applications use it.

Initially, SQLite emerged from the need to ensure reliable database operations on the USS Oscar Austin destroyer.

It is supported by _three people_. They do not admit external contributors. You can’t just send a pull request and hope the patch will be approved. SQLite’s development is backed by the company Hwaci. Interestingly, they seem to be involved in music too. Their website is extremely succinct. Office in a private house.

For each line of SQLite code, there are over 600 lines of test code. Tests cover 100% of branches (and 100% MC/DC) of the library. The test suite is highly diverse, including fuzzing tests, boundary value tests, regression tests, and tests simulating operating system crashes, power failures, I/O errors, and out-of-memory errors. Originally, SQLite was an extension of Tcl, and its main test suite is written in Tcl.

Interestingly, some SQLite tests are proprietary. The TH3 (Test Harness 3) test suite, which provides one hundred percent code branch coverage, is proprietary, and access to it is restricted.

I know of no other project that is open-source yet has paid tests.

To gain access, one must join the SQLite Consortium with an annual membership fee of 120 thousand dollars.

This is an interesting business model. Developers generate profit through licenses, paid support, service maintenance, consortium membership, and commercial extensions.

SQLite does not have a Code of Conduct (CoC); instead, they use a Code of Ethics derived from the “kinds of good deeds” from chapter 4 of the Rule of Saint Benedict. Google SQLite Code of Ethics, it’s a very intriguing document.

At the beginning of all source files, instead of legal notices, there is a blessing: “In place of a legal notice, here is a blessing: May you do good and not evil…”

SQLite creator Richard Hipp (D. Richard Hipp, DRH) couldn’t find any version control system that suited him, so he created his own called Fossil. Of course, Fossil is based on SQLite.

It reminded me of how Linus wrote Git.

DRH also wrote his own parser generator called Lemon.

All the above is described in more interesting details (and expanded) in an article on Habr, which is a translation into Russian of an article by Avinash Sajjanshetty. For originals — check the comments

Discoveries Underwater, Online, and Under Anesthesia | January 16 2025, 22:18

Some novelties for me. Sharing, as they’re likely new for you too. About bulbous bows on ships, about how the body of water under the Paris Opera from “The Phantom of the Opera” actually exists, about how 70% of all global internet traffic passes by my house, and about how no one really knows how general anesthesia or paracetamol works.

= First Discovery =. Large ships may have this thing in front, called a bulbous bow. I didn’t understand what it was for, it seems to disrupt hydrodynamics. Turns out, it’s quite the opposite. It improves it. It creates a counter wave that partially offsets the main bow wave occurring when the ship is moving — especially at high speeds and for large vessels. Each ship’s bulb is designed individually.

= Second Discovery =. Remember “The Phantom of the Opera”? There was a body of water under the Paris Opera where Erik and Christine sailed. Well, it’s not fictional. There really is a body of water beneath the Paris Opera House.

The opera house required a much deeper basement space beneath the stage than other types of buildings, but the groundwater level turned out to be unexpectedly high. In February 1862, wells were drilled and in March eight steam pumps were installed, which operated around the clock, yet the site remained damp. To solve this problem, Garnier designed a double foundation to protect the structure from moisture. The design included waterways and a huge concrete tank (cuve), meant to reduce the pressure of the external groundwater on the basement walls and to serve as a reservoir in case of fire. The construction contract was signed on June 20. Soon, a persistent legend arose that the opera house was built over an underground lake, inspiring Gaston Leroux to include this idea in his novel “The Phantom of the Opera”.

= Third Discovery = Of course I knew it was a lot, but today I saw the number — 70% of internet traffic passes through a place just about 10 minutes from where I live. Around me, more than 130 data centers have been built covering an area of three million square meters. Quite a thrilling neighborhood. I hope no ill-wishers will mark this place on the map with a cross. This cluster specifically consumes about 150 TWh, which by my estimates is like ___THREE MOSCOWS___.

= Fourth Discovery = It turns out that no one really knows how general anesthesia works. There are theories in Wikipedia on how anesthesia functions. But though they exist, none of them is completely convincing. In 2019, it was found that human-like anesthesia even works on plants, like on a mimosa and on a venus flytrap—they stop responding to touch under it. Not to mention, as everyone knows, plants don’t have a central nervous system at all. Not a brain, not even neurons. There’s also xenon anesthesia. Meaning, it’s fundamentally an inert substance, doesn’t react with anything, but somehow works. They recently figured out how aspirin works, but there’s still no full understanding of paracetamol.

Exploring the iBot: A Leap in Personal Mobility Technology | January 11 2025, 00:34

Today I finally made it to the exhibition Paris 1874: The Impressionist Moment – over an hour in line. Nadya says – look at that interesting wheelchair.

Nadya and I had been wheeling her dad around for many years. He fell ill, ended up without legs, and the wheelchair literally became an extension of him in everyday life. Just your “basic” wheelchair. We took him to the Black Sea by car several times, traveled abroad, and of course experienced all the “delights” of accessibility in Russia. Actually, I don’t even know how to translate “accessibility” into Russian properly. “Barrier-free environment”? It was not barrier-free, it was downright threateningly barriered. For example, we simply couldn’t get to the doctor at the clinic because the elevator was too narrow.

So, back to the wheelchair in the photograph. It’s the iBot by Mobius Mobility. It’s no longer customary to call them “wheelchairs,” it’s a “personal mobility device.” It operates both in balancing mode, where only two wheels are used — essentially, like a Segway, and in a four-wheel mode when terrain handling is needed. In two-wheel mode, the seat raises to a height of 91 cm, allowing the seated individual to be more or less at eye level with others. It also has a mode for climbing/descending stairs — that’s really cool. The range is 35 km in balance mode.

The iBot was invented by the same guy who invented the Segway — Dean Kamen. Overall, this startup is several years old, with several generations of the iBot already released.

Yes, the price of such a device ranges from 32 to 40 thousand dollars. Quite a lot. But it’s said that about half can be covered by insurance (still a lot, though).

There are a few people around us in wheelchairs. And there were few in Russia too. But in Russia, there were few because they just couldn’t leave the house, and each outing for the family or companion turned into a project that would be good to plan in advance. Here in the USA, there are few for another reason. Here, in case of problems, they don’t just hand out a wheelchair, they give a new leg or a new joint. And only if it’s completely irrepairable, then they resort to a wheelchair. Yes, such operations can cost a fortune for people without insurance, it’s a known problem, but usually, some charitable foundations are found.

For example, near our house is the organization ECHO (Every Citizen Has Opportunities). Besides helping financially with adjusting to a new way of life, they offer jobs to people who land in such trouble, they also provide free transportation (to work, to the store, etc.) and socialization.

Exploring the Boundaries of AI in Dreyfus’s Pioneering Work | January 10 2025, 01:40

Currently skimming through a book, Dreyfus (1972) – What Computers Can’t Do – The Limits of Artificial Intelligence. In it, across 300 pages, the author convincingly, with numerous references to scientific papers, argues that, for example, programming a chess game is impossible, and intuitive and situational human tasks, such as understanding natural language, are even more profoundly unprogrammable.

The conclusion of the book is that instead of striving for complete autonomy, AI researchers should focus on enhancing human intelligence and exploring the fundamental differences between human and machine minds. They should probably read this book first.

And 53 years later, I am using AI to translate and extract key ideas from this book.

Hubert Dreyfus passed away 7 years ago. Overall, he probably began to suspect long ago that things were not as he had written in the book, because in 1992 he wrote a second series “What Computers Still Can’t Do”.

But the funniest thing is that the 1972 book was printed in Russian in 2010 and can be purchased; it is still widely sold on “Ozone” for 976 rubles. Labeled as NEW!