Occam’s Razor 🪒

.. is a problem-solving principle stating that when faced with competing hypotheses, the simplest explanation requiring the fewest assumptions is usually the best.

The "razor" refers to shaving away unnecessary assumptions or "entities" to get to the most straightforward solution.

Ref.: https://www.britannica.com/topic/Occams-razor

Image ref.: GPT Image 2.0 👀 (good 👍)

Круто расписано 😁

​​We’re not saying that registration for the Proggy-Buggy Towel Contest is open.

But if you keep seeing 42 everywhere…
maybe it’s a sign.

13 problems.
42 minutes.
One towel.

Прикольный - контест ProggyBuggy: на 42 минуты дается 13 задач на логику -> будет 6 июня. В Алмате, Астане и Караганде оффлайн локаций есть.

On June 10, 2026, the Kazakhstan Stock Exchange (KASE) invites to a workshop dedicated to investing in AI and practical application of artificial intelligence.

⭐️ The event will be held in collaboration with Silkroad Angels, a closed club of Central Asian business angels investing in Silicon Valley tech startups.

During the workshop, we will discuss:
✔️ AI as a key technological trend and driver of the global economy;
✔️ specifics of venture capital investments in AI startups;
✔️ approaches to project selection and investment risk management;
✔️ access to deals and the Silicon Valley startup pipeline;
✔️ practical application of AI in business;
✔️ main AI trends of 2026 and real-world implementation cases.

🎤 Speakers:
Abai Absamet, co-founder of Silkroad Angels, an angel investor community at the Silkroad Innovation Hub in Silicon Valley;
Arman Suleimenov, serial entrepreneur, founder of Zero To One Labs and nFactorial School.

👥 Participants will also enjoy a networking session, Q&A and the opportunity to discuss current AI and venture market trends with experts and investors.

🗓️ June 10, 2026
📍 Almaty, 280 Baizakov St., Almaty Towers BC (North Tower), 1st floor, Press Room.

📌 Registration link is via this link.

https://luma.com/x4hpxhhq

Build Small Hackathon by Gradio & HuggingFace 🤗

The premise:
≤32 billion parameters.
Two tracks.
Two weekends (May 29 - June 8)
$15K in prizes.

The first track: Backyard AI
Judged on whether the person you built the app for actually used it. That's the bar. Not a leaderboard. Not a benchmark. A real human in your life who got measurably more help from your software.

The second track: An Adventure in Thousand Token Wood
For building things that are weird, joyful, and wouldn't exist without AI.

Registration just opened and closes May 27.
❤️‍🔥 Whether you've been to a hackathon before or this would be your first, come build something small with us. Read more.

https://huggingface.co/spaces/build-small-hackathon/registration

🌟Spectral::Cup 2026 Round 2 (Codeforces Round 1100, Div. 1 + Div. 2)🌟 starts on 23 May 2026 at 14:35 UTC.

It is rated and open for everyone!

Round will be held with the support of Spectral::Technologies, an HFT fund building trading strategies and low-latency infrastructure for global markets.

🎁 Prizes for Round 2:
Laptop cases and Codeforces-inspired anti-stress toys.

It is the second round of the Tournament. Don’t miss the round!

Please, join by the link https://codeforces.com/contests/2229

The Problem with "Trust Me Bro" Benchmarks

* The "Hype" Loop: Models are often specifically optimized to pass popular benchmarks, meaning they learn the answers rather than solving the underlying logic.

* Contamination: It is suspected that testing datasets are included in the training data, leading to artificially high scores.

* Marketing > Science: Many, such as the SWE-Bench Pro comparisons, are used to boost investor sentiment rather than providing transparent data.

* Lack of Evidence: Often, these benchmarks lack the underlying scripts or raw outputs needed to verify the claims.

Image credits: OpenAI reddit / ~ text Gemini

«Онлайн-кодирование» в контексте машинного обучения (ML) — это процесс преобразования категориальных признаков в числовые значения непосредственно в процессе обучения модели, не требующий предварительного полного сканирования всех данных (batch-обучение).

Онлайн-методы кодирования, такие как целевое кодирование (target encoding) или хэш-кодирование (hashing trick), позволяют кодировать новые данные с минимальными затратами ресурсов.

Attention Is All You Need = Назар аудару - саған керек жалғыз нәрсе

* Deep Learning - терең оқыту;
* Dataset - дерекқоржын;
* Vanishing Gradient - жоғалып бара жатқан градиент;
* Exploding Gradient - жарылып жатқан градиент;
* Softmax - жұмсақ максимум;
* K Nearest Neighbors - K жақын көршілер;
* RandomForest - кездейсоқ орман;
* DecisionTree - шешуші ағаш;
* Pooling Layer - біріктіру қабаты;
* Hidden Layer - жасырын қабат;
* Dropout - тастап кету;
* Multi-headed Attention - көп-басты назар;
* Multi-armed Bandit - көп-қолды қарақшы;
* Backpropagation - кері таралу;
* Weight Decay - салмақтың ыдырауы;
* Loss Function - шығын функциясы;
* Chain-of-Thought - ойтізбек;
* Feature Extraction - белгісуыру;

#ml_terms_in_kazakh