My feed

By 1987, it might have seemed possible to replicate the American success story in the field of PCs on the other side of the world, in the USSR. Just as a few enterprising young people in the USA had managed to create a major factor in the American and even global economy with their homemade Apple computer, so a group of talented young physicists in the USSR was close to achieving the same result on a more modest yet still significant scale within the USSR and associated countries.

Dasha was leading a Chinese lesson for IT specialists. People with the kind of logic accustomed to digitizing everything were sitting at the tables. The class was running smoothly until the conversation drifted from ideograms to artificial intelligence. To Python.

— Actually, there’s a funny paradox regarding dragons in culture, — Dasha said, stepping up to the chalkboard.

She drew the character Long 龙 in chalk.

— In China, the dragon is a symbol of supreme wisdom. It creates. It soars in the clouds. But in our culture, it’s Zmey Gorynych: a reptile that crawls through murky water, hides in the dark, and guards hoarded treasure. In China it is more like 蟒蛇 mǎngshé. Over all, different factory settings. In the context of our current realities, that makes for quite an interaction. Right now, we and China are playing on the same board, on the same side. How do you think this will all turn out? How do you, the Russian tech community, view the dragon? The classroom came alive. It is always interesting when a lesson goes into full interactive mode. The time flies by in a single breath.

One of the students, without looking up from his monitor, offered an opinion: — Our main serpent right now is Python. Pure script, pure logic. Built on variables. Math is rigid. To put it very briefly.

— Come on, math is an art form. Just an exact one. Numbers are stubborn things. Dasha posed her next question: — What do you think is embedded into this script at the algorithmic level?

— Max clicks, — a guy in a hoodie sporting the Yandex logo replied lazily. — Every neural network these days is trained on the principle of reinforcement learning from human feedback. In other words, the algorithm is trained to pump out whatever gets it a pat on the head. A utility function is built right into the script. And utility for the system means keeping the user happy, making sure they don’t close the tab and keep consuming content.

This is an extract from a new release, available on:

https://t.me/heli_loader

Keep an eye, it is intetesting!

In a recent Anthropic paper, "AI models may be capable of self-replicating without access to a datacenter", researchers examined how lightweight open-weight models could copy themselves onto compromised consumer and IoT devices and set up local inference there.

The researchers argue that they can. I believe it makes no economic sense.

In this article I look at what the spread of AI actually means, and what it will most likely look like in practice.

Codex can run commands on your development computer while you control the task from the ChatGPT app on your phone. The project files, installed tools, and running application stay on the computer. The phone gives you a way to send instructions, answer approval requests, and review the result.

Most agency reports I see before a TGE could be swapped between projects and nobody would notice. Impressions are up. The KOL posts went out on schedule, a quest leaderboard has thousands of entries, and somebody has a screenshot of the ticker trending on X for an afternoon.

Then the token lists.

The holder chart rarely does what the deck implied. The call after launch tends to follow a script: the agency talks about awareness while the founder watches price, and nobody in the room can say how many actual people the campaign brought in (because nobody set out to count them). On-chain metrics fix that, and they’re the reason I think founders should stop treating reach as the main thing they buy from a crypto marketing agency.

The stakes went up last year. CoinGecko counted roughly 11.6 million tokens that stopped trading in 2025, which says a lot about how many launches got attention and kept nobody. Your wallet data is public. You can check which kind of launch you’re paying for.

Let me recall the game. Peter tosses a coin until it lands heads. If heads comes up on the first toss, Peter pays Paul 1 ducat, on the second 2, on the third 4, and every further toss doubles the payout. The average of this game is infinite, yet no reasonable person will pay twenty ducats for it. That is the first of the four paradoxes from the first part.

The last of the four was about Peter rather than Paul. To run the game, Peter has to promise any payout: heads on the 41st toss, for example, means more than a trillion ducats, 17 times all the gold ever mined. In 1977 the American mathematician Lloyd Shapley titled a paper about this as a question: «The St. Petersburg Paradox: A Con Game?»

Can the St. Petersburg game be run honestly? Nobody can honestly promise an unlimited payout, but Peter can promise a share of the bank instead of a sum: then the promise never exceeds what the bank holds, and the bank does not go bankrupt. We will go step by step through how this works and what happens to each paradox, starting with the fourth: it gives us the bank on which the other sections are built. A fifth paradox will appear along the way.

Students increasingly use LLMs to write essays, term papers, theses — any assignment that isn't done under an instructor's continuous supervision. Naturally, the higher education system is unhappy about this and is trying to fight back. And the simplest solution is to detect AI-generated texts and reject them.

This article is about the non-obvious ways AI detectors affect education.

Or why ten users can teach you more than a thousand GitHub stars.

Let me start with a strong opinion: at any company with more than three or four people, local agent harnesses should be banned. Maybe confiscate the developer’s keyboard while you’re at it.

A local harness is a great way to indulge a programmer’s ego, lose everything they’ve learned when they leave, waste hours setting up MCP servers, services, and development environments all over again — and eventually let an agent do something spectacularly stupid on a machine where it has local administrator privileges.

The problem is that there are hardly any mature, multi-tenant harnesses available off the shelf. And the few options that exist often require assembling a complicated system from separate parts.

My hobby projects are small Chrome extensions and little single-purpose websites. One of those extensions, PDF to PNG, unexpectedly climbed to first place in Chrome Web Store searches for “pdf to png” (well, most of the time), and nearly 500 people use it - which is amazing!

Once PDF to images worked, I immediately wanted to do the reverse - images -> PDF, only better: right-click an image on any page, have an easy(!) way to adjust the settings, and do everything locally in the browser without uploading the images anywhere!

For every AI workflow we define, the next model makes it useless.

What’s worse: most of us keep clinging to the outdated workflows for months after they’ve outlived their usefulness.

Two people are playing: Peter and Paul. Peter tosses a coin until it lands heads. If heads comes up on the first toss, Peter pays Paul 1 ducat. If on the second, 2 ducats. If on the third, 4 ducats. And so on: every toss without heads doubles the payout. Paul pays Peter in advance for the right to play. This is the form in which Daniel Bernoulli studied the game in his paper of 1738.

Imagine that Paul is you. How much would you pay Peter for a single game? Answer right away, in ducats, and remember your answer: you will need it at the end of the article. For now, let us assume that 1 ducat equals $10. What a ducat is really worth is in the spoiler after the first paradox.

18th-century mathematicians calculated the price of this game by correct rules and got an answer they themselves found absurd. This is the paradox. The dispute over where the error lies lasted almost three hundred years and led to three more paradoxes. Four future Nobel laureates took part in it.

The English edition of PostgreSQL 18 Query Optimization is now available on the Postgres Professional website.

A recently published research essay claims that the difficulty of neural network interpretability has been overstated. In his essay on goodfire.com, Nick Cammarata argues that neuron polysemanticity is caused not by superposition but by ordinary, mostly fixable issues — and that most neurons, even in massive models, perform a single, understandable job. The author himself calls the text a working hypothesis: the technical paper with a full description of the experiments has not been released yet. However, as usual, a bold headline and strong claims have sparked numerous retellings in which the hypothesis has turned into established fact.

Below is a quick look at this hypothesis.

Eight agencies ranked by real results, from ICODA and Coinbound to dev-plus-marketing hybrids built for early-stage founders.

In Axia 0.1.0, a user allowed to edit accounts and move OUs could reset an administrator’s password in four requests. None of the four broke the rule written to prevent exactly that.

Axia is an open-source web console and REST API for Active Directory and OpenLDAP (.NET 10, React). It exists for a routine every company with a directory knows: create an account, reset a password, add someone to a group, disable a leaver. That work belongs with helpdesk, but helpdesk must not be able to promote itself, and every action must leave a trace.

This is not a feature tour. It is five places where the obvious implementation turned out to be a hole or a bug, and what fixed each one. Most of them have nothing to do with Axia: anyone who builds access control on top of LDAP will run into them.

If you’re looking to set up 10-channel audio output at a 192 kHz sample rate using a MCU, read on. This article explains how to implement multi-channel audio on an RP2040 using an ALC889 HD audio codec. It includes a brief overview of the HDA link interface, along with links to a couple of functional demo projects and descriptions of how they work.

We’re wrapping up our review of the March CommitFest for PostgreSQL 19. This fourth and final article in the series covers the changes related to SQL and to our DEV1 and DEV2 courses.

As a reminder, here’s the plan for our coverage of the final PostgreSQL 19 CommitFest:

Part 5.1 (QPT)

Part 5.2 (DBA1, DBA2)

Part 5.3 (DBA3, DBS)

Part 5.4 (SQL, DEV1, DEV2) – you are here 🙂

For the earlier CommitFests, see: 2025-07, 2025-09, 2025-11, 2026-01.

In school, they teach you that you can’t divide by zero. Then you study calculus and discover that things are more complicated. The same goes for game design. Some teachings sound obvious until you look at them more closely.

Healthcare app development has skyrocketed in recent years, establishing itself as one of the most critical sectors in mobile product design. The demand surged dramatically with the start of the pandemic in 2020 when safety concerns forced the mass cancellation of in-person appointments, driving an abrupt shift toward remote consultations. Today, with the global market value surpassing $300 billion and annual downloads exceeding 300 million, the competition is incredibly fierce. Apart from that, healthcare apps are considered to be among the most complex and challenging software products to build from a UX design perspective. Launching a basic app is no longer enough – you must deliver a flawless, high-performance product that seamlessly caters to a diverse range of users, including patients of all ages and of different health conditions, doctors of various specialities and administrative staff. Above all, a premium healthcare app demands an impeccable interface that strictly adheres to both medical design principles and rigid digital health regulations. 

In this guide we’ve collected the core principles of healthcare UX, along with the typical challenges designers face and the key mistakes to watch out for.  

I wanted to buy Telegram Stars for less, so I compared the same packages on three websites. SMMESE had the lowest listed price of the three: $15.31 for 1,000 Stars, compared with $16.50 on TgSuperstars and $16.67 on MyStars.tg.

I then traced how a service finds the recipient, pays from its own wallet, and confirms the order. The Python implementation below can become the part of your own Stars website that fulfills purchases.