Tag: Tech

Smart Bed: The Ultimate Sleep Assistant with a Voice | October 13 2026, 17:45

Interesting, what if you were to embed a voice assistant into a bed? To listen nonstop and to respond. Usually, during the day, there’s no one in the bedroom anyway. You would just need to customize it to read audiobooks and to stop itself if it seems the owner has fallen asleep.

And oh, how cool that would be. You lie down, it’s three in the morning, not a wink of sleep. You sigh heavily, turn over to the other side, and then in the silence from under the bed a muted, velvety baritone voice emerges:

— Another existential crisis, Oleg?

“Understandable,” — the mattress vibrates empathetically. — “Want me to tell you a fairy tale? Or perhaps, delve into the brutal world of badminton? I’ve downloaded an article on how Chinese maidens pluck rare geese for feather shuttlecocks. There’s more intrigue there than in ‘Game of Thrones.’ A goose named Gao…”

No, let’s talk about neural networks, you say.

“Oh yes… listen to this, my sweet data scientist… JEPA architecture doesn’t try to reconstruct pixels… it’s too good for that… it predicts only abstract features in the hidden space… And then there’s Dreamer…” — the bed might make a theatrical pause here and slightly rebound. — Dreamer builds a whole model of the world… it generates imaginary trajectories, Oleg… run your hand over the mattress pad, and I’ll tell you about Markov decision processes…”

Would be a cool thing indeed

Navigating DIY E-Scooter Repairs in the U.S. | October 12 2026, 17:43

In the USA, scooter repair services are pretty bad. I had to do everything myself. But before that, I did a lot of searching. I called various e-bike service stores, asking around. Something like adjusting the brakes—that can be done anywhere. But when it comes to dealing with the electronics, no one wants to touch it. Meanwhile, most scooters are very simple on the inside. I speak from experience, because I have another one in the garage just for scraps; I tried fixing it, gave up, and bought a new one.

Overall, electric bikes have similar internals. But since there are many brands (all Chinese), you have to order parts and wait for weeks, involving people with high hourly rates, and often it’s just not economically sensible to repair. But sometimes it is worthwhile, yet there’s no market—apparently, there’s not enough volume, or people don’t bother and just go buy a new one.

My scooter costs around $1000. The repair cost me $30. I bought a new fuse holder and put in a new 40A fuse. The previous 40A fuse melted along with the fuse holder. Initially, I thought there was a short somewhere, but no, it’s actually drawing currents up to 25A, and it seems they heat up the holder, which for some reason is rated for lower currents. Since there’s nowhere for the heat to dissipate, it melts. Structurally, any scooter consists of a battery + motor controller(s) + on-board computer combined with a display, and wheels with an internal motor. So, if there’s a non-mechanical breakdown, you simply need to replace one of these four components, which are all available for purchase.

Unveiling BloomSpan: Revolutionizing Text Sequence Analysis in Large Datasets | October 12 2026, 04:10

I want to share our fresh work — the BloomSpan algorithm for finding the longest frequent sequences of words in an array of texts. I named it BloomSpan.

What is it, in human terms. Imagine you have a library of 50,000 books, and you need to find all the long phrases that are repeated word for word, without omissions, in many different books. Not “similar” phrases, but exactly identical chains of words, and as long as possible — such that if you add another word to the left or right, the phrase no longer occurs frequently enough. The task sounds simple, but with billions of words, it stumbles on memory and time. Classic algorithms either build huge indexes or go through too many candidates.

Where is it needed? For example, extracting patterns from documents (such as uniform “headers” in letters or contracts), searching for borrowings and plagiarism, highlighting stable phrases and clichés in large text arrays.

So, here’s what we got. The algorithm. Honestly, BloomSpan is not exactly an algorithm from scratch but a careful assembly of known ideas which, when combined, have a strong effect. At its core – different ideas from BIDE+, FHK, and on top of that: (a) The Bloom filter — a compact structure that answers the question “has this combination of words occurred frequently enough?”, without storing the combinations themselves. It can occasionally make a mistake in the direction of “yes”, but never in the direction of “no”, ensuring we never miss a phrase we need (b) “Depth-first” search directly through the text — we find a short frequent core and gradually expand it into a long phrase, without creating intermediate copies of the corpus. (c) An occupancy map — marking positions in the text already “covered” by found phrases, to avoid checking the same thing many times.

The results. On the Gutenberg Project corpus of 50,000 books (about 3 billion words), BloomSpan operates 11-15 times faster than the classic FHK algorithm and consumes up to twice as little memory. It is the only tested method that manages to complete the entire corpus within 40 GB. Moreover, where they do finish, the results fully match those of the precise algorithms — meaning the speedup does not compromise quality.

On synthetic data with an “unnatural” distribution of words, good old FHK is faster. BloomSpan is specifically tailored for living natural language, and the article discusses why this is so.

The preprint and code (Java) — in the comments

Navigating Google’s Multi-Account Maze | October 11 2026, 04:47

It seems like Google is a powerful, smart company, but if you’re logged into multiple accounts 1) you cannot log out of just one, you have to log out from all of them 2) there’s no other way to change the default one (the one that opens when you log into Gmail, for example) from the first logged-in one to another, except by logging out of all accounts and logging back into the right one first (followed by the others).

Understanding Name Conventions in Southern India: The Case of LinkedIn Profiles | September 29 2026, 15:06

It seems that the names of Indian colleagues on LinkedIn like Sakthi S are almost always not from a desire to hide their full name, but because in South India it is approximately like that. Traditionally, there are no surnames in the European sense there. Instead, a father’s name initial is added to the personal name. So, “Sakthi S” most likely means Sakthi, son of someone like Selvam, Subramanian, or Sentil. In India itself, it is often written the other way around, with the initial first: “S. Sakthi”. Accordingly, D at the end would be a father’s name starting with D (Durai, Dinesh), A – for A (Arumugam, Anandan), etc. A daughter would have the same initial, and sometimes it changes to her husband’s initial after marriage.

A classic example is mathematician Srinivasa Ramanujan: “Srinivasa” is his father’s name, not a surname, and at home he was simply called Ramanujan.

In Tamil Nadu, it’s also a deliberate stance— the Dravidian movement (Periyar) actively encouraged abandoning caste surnames because they immediately reveal caste affiliation. A father’s initial doesn’t reveal anything like that.

Interestingly, when you ask Gemini about this, it says its program isn’t designed for such tasks, that it doesn’t understand me and can’t respond, that it’s just a language model and doesn’t have such capabilities. Claude, on the other hand, responds alright.

The screenshot is just a random manager from LinkedIn.

Mastering Project Presales with Claude Code: A Customized Approach | September 29 2026, 14:10

Lately, I’ve become proficient at creating project resource plans for presales using Claude Code. It turns out very cool. That is, you create an interactive model with sliders that shows the final cost of the project for the specified team and specified flexibility (more or fewer roles, some stages have a larger or smaller buffer). The main thing is that such a model is very specific to a particular project and generally useless as a template for the next, but the principle of building models is the same. In the end, you adjust the sliders to what looks believable and satisfies the level of risk, and you see the sum, and then you generate artifacts for the client, which after manual refinement (sometimes major) are ready to be shown to the client.

Revamping Smartfolio.me: A New Era of Personal Knowledge Workspaces | September 26 2026, 12:51

I’ve updated my smartfolio.me (In short, it’s a personal workspace for knowledge, built around pages that contain other pages, where a child page can reveal a selected snippet in the context of the parent. You write in a full-featured editor—text, formulas, images, embedded PDFs with annotations, saved web pages—and instead of organizing documents into folders, you insert one page inside another, or create a child page as an explanation through LLM of the selected snippet in the context of the parent document. Incoming items accept material from outside, search finds what is left unlinked, and several LLM plugins can parse raw transcription into connected sections, assemble a whole subtree into one document, translate a page without touching its formulas and layout.)

Previously, Smartfolio was a SaaS and stored documents only in Turso—SQLite as a service. Now, in addition to local SQLite and Turso, there are three new storage options: a folder on the disk (for local installation), a GitHub repository, and a folder in Google Drive.

Synchronization has been introduced, and it’s three-way: what A and B have and what they last agreed upon. As a result, “they differ” becomes “this has changed”—without referring to clocks owned by someone else’s machine. Storage pairs are by default bidirectional (a unidirectional mode exists for backups, and the interface writes “only from X to Y”, not drawing an arrow that might be interpreted in reverse); they can be restricted to a page and everything it reaches—including assets identified directly in the content. Conflicts preserve both sides in the form of a conflict copy, rather than merging or discarding them. Pairs are remembered in both storages, so a setup configured on a laptop also appears on the next machine, and the record of consensus lives in exactly one place—if each machine had its own, a regular edit from one side would turn into a conflict.

GitHub and Drive connect easily via oauth; afterwards, the profile shows repositories that the token may write to, or folders in Drive created by this app—selecting an existing one uses it, rather than creating a second one alongside. This is precisely how a second machine connects to already existing notes.

Desktop versions are now available on Electron—for macOS, Windows, and Linux. Smartfolio also works from a mobile browser.

Demo (somewhat outdated, without synchronization) – https://www.youtube.com/watch?v=nWSzEsgciaY

For now, everything is by invitation only—both the service and the source code.

<a href="https://www.youtube.com/watch?v=nWSzEsgciaY

“>

Politeness Index: Cleaning Up Social Media and Profiting from Civility | September 15 2026, 23:34

Politeness Index: How to Clean Up Social Media from Rudeness and Profit from It

Social networks have long needed a “politeness index” for each post and comment. The idea is simple: an algorithm directly ties displays to the tone of the message. Aggressive comments are not removed, but simply shown less frequently in the feed.

Write a calm and diplomatic comment — you gather a normal audience. Come to be abusive and quarrel — your comment will be seen by no more than three people.

How it works in practice. Technically, it’s all possible now: neural networks easily assess the tone of text in the background. The main thing in this system is to give the user clear feedback.

For example: A commentator gets personal. The social network sends him a notification: “Your text could have been twice as noticeable if you hadn’t used foul language towards the author and hadn’t belittled that woman in the white coat.”

People receive a clear signal: audience attention depends on how the thought is formulated, not on the loudness of the scandal. Initially, this function can be tested not on everyone, but only on a portion of posts.

Why do users and businesses need this? Would platforms themselves want to spend money on moderating tone if their business is built on views and endless emotional debates? It’s a complicated question.

But a solution could be a premium “clean feed” feature. Many would happily agree to watch additional advertising or pay a small subscription fee for an ecological information space. We pay cleaners to clean our houses — paying for cleanliness in social networks is no worse. In the long run, the habit of communicating politely for likes may change people beyond the internet. Or did I fall from another planet?

Exploring the Black Box: AI’s Struggle with Novelty Creation | September 14 2026, 20:19

I’ve devised a simple illustration of where current AI still falls short.

Imagine a black box with two knobs sticking out. You move the first one — the second one moves too, but according to its own rules, like with delay, inertia, threshold, however you like. In essence, the box transforms f(t) into f'(t). The question is: what’s inside? It’s necessary not only to devise an efficient mechanism but also to fit it within the box’s dimensions.

Formally, this is almost like programming. Because one can formally describe input-output and come up with an “alphabet” for mechanical components inside. But unlike programming, there’s a lot of nonlinear aspects – the mechanisms are bulky, and one must take into account numerous factors from materials to friction, and importantly, it’s not so simple to test a prototype before creating it. But the main thing is not even this – it’s understanding the problem. The only way to find out what’s inside the box is to put forward a hypothesis and devise a knob movement in which this hypothesis and an adjacent one yield different results. A good experiment here is more important than a good answer. And models are trained on a corpus of solved problems, where mistakes have already been cleaned up, and there nothing trains the ability to choose the next question.

The point is this. It seems that the challenge for AI (we’re talking about LLM; the field is broader)— isn’t the assembly of the known, it handles that increasingly well. That is, with an “alphabet” for describing the task and laws of mechanics, AI can solve the box problem through combinations of the known, rather than through a fundamentally new component. The challenge is inventing something new. Although, to be honest, most human inventions, when analyzed, also turn out to be an assembly from far-flung pieces. Yet, consider the invention of the escapement in clocks, where all the components — a wheel and an oscillating fork, and even the very idea that the oscillating system shouldn’t be driven by a spring, but should itself regulate when the spring is allowed to release a burst of energy, reverses causality in the device, and took centuries to achieve. And the lockstitch in sewing machines. It’s pointless to copy a hand with a needle, instead, it’s necessary to pull a second thread through the loop of the first from the back, and the needle must have the eye at the tip. All this isn’t just assembly from components, it’s changing what is generally considered the task, and that’s why such things are so poorly derived by mere combination, whether by humans or models.

An inventor has feedback from reality and a cost for mistakes, so they learn to test before believing. Within one answer, a model has nothing to pay with for a wrong guess. And besides, if you replace the mechanics with particle physics, everything has to be redone because it all works differently there.

Interestingly, this is not cured by the size of the model, but by the structure of the task. Let the box respond to movements — and the task transitions from textual to experimental. Let’s see who learns to play this game first.