Sami Badawi

Yesterday @armanbil.ge from the Typelevel Foundation gave an amazing presentation about #CatsEffect for #Scala. There are a lot of speed improvements since v3.0.0. He explained how it works under the hood, and compared it to Java's project Loom.

Top Moda

I have been listening to the Big Think Interview. They have high quality interviews with scientists and other big thinkers. For example: The science of stress, coping, and recovery, in 22 minutes | Lisa Damour Here on YouTube: www.youtube.com/watch?v=q57f...

I expect a wave of prompt injection attacks. You should understand the mechanics before you get hacked when using AI. embracethered.com/blog/posts/2...

Squeezed Devstral Small 2 24B, Mistral's 24 billion parameter LLM, onto my 32 GB laptop. I gave it a Raylib code test. It handled the test in C, C++ and Rust. Also made graphics demos in JavaScript and ported to TypeScript. Capable, but huffing and puffing. 😮‍💨

Switched to llama.cpp and OpenCode for self hosted coding LLMs. On a 32 GB RAM laptop I get decent speed and a 100k context window. Models that fit here are only good for pair programming, so speed is important. llama.app

I tried self hosting open weight models / coding LLMs on my modest laptops for a week. Are there any use cases for this now? Yes. If you are offline, broke or cannot send code over the internet for privacy or legal reasons. But be prepared to pair-program with your LLM.

Last year AI = ChatGPT That level of industry dominance scares me. So I self hosted 3 coding LLMs, from 3 continents * Gemma 4 E4B 6 GB * Qwen3.5-9B 6 GB * Devstral Small 2 24B 9 GB They all did real work on my modest laptop. For free 🤯 Update my post blog.samibadawi.com/2026/08/self...

Elusive eclipse sunset

Yesterday I spent hours removing vibe coded verbose documentation and mock unit tests. 🤖 There is a fine line between best practice and slop. 🫥

My coding LMM triggered AI safeguards shutdown. 🤯 I am coding a database insert to a table called audit. Not doing cybersecurity work. 🙁 Guardrails are good, but I thought they were more sophisticated. 💂 (Gemini suggested to temporarily rename audit table to log_history)

Tried Self Hosting Coding LLMs. Setup is easy with LM Studio, OpenCode and Gemma 4 E4B. It worked on my MacBook and PC laptop running at good speed. Gemma from Google DeepMind handled my code test questions without web search. 👍 My cleaned up blog post: blog.samibadawi.com/2026/08/self...

How Apple Won The AI Race By Doing Nothing * $100 billion data center spend is a race to the bottom. * LLMs are commodities. * Apple built hardware where you can run local models. Contrarian view on AI from Feb 2026: www.youtube.com/watch?v=401S...

I have tried #OpenCode. Good ergonomics, similar to Claude Code. In OpenCode you always see the number of tokens and money spent. Your LLM does useful work, but it also burns tokens by reading through long logs. Humans 😵‍💫 assume LLMs 🤖 would warn us. In OpenCode you notice problems fast.

#OpenCode, an open source AI agent harness that runs in a terminal, has momentum. I liked minimal cursor based GUI and connection to many LLMs, not just from Anthropic, Google and OpenAI. Giving some protection against vendor lock in and rising cost. www.youtube.com/watch?v=KvFm...

This video gives a peek under the hood of AI agents. It demonstrates new tools in 4 agent categories: 1 No code 2 Low code 3 Agent harness 4 Full code www.youtube.com/watch?v=ZvDk...

Frigjort ( Liberated ) by Rudolph Tegner, 1900

My #coffee habit: ☕ to wake up ☕ get that boost ☕ combat afternoon slump Last week I only drank morning coffee, and had no more afternoon slump. 🤔 ☕ is advertised as an energy boost, but it's an energy loan.

#GoogleIO A 10x moment What happens when we scale X, Y and Z 10 times? Great questions. Most answers were: "We don't know yet." www.youtube.com/watch?v=2n41...

Herb Sutter on reflection in C++26 at CppCon 2025. I tried some reflection examples on GCC 16 on Linux. Simple to use and worked. 🤩 www.youtube.com/watch?v=7z9N...

Can you find what you miss in C, inside C++, without all the complexity? Tom suggest only using: * Methods in struct * Namespaces * Generics He calls it: C with generics. It only uses C++ features that work on microprocessors. I tried it, and was impressed 🤩 www.youtube.com/watch?v=ur-b...

Tried the new C bindings in new Zig 0.16. #Zig is doing translation from C during compile time using a new build function: addTranslateC. It was simple to make direct bindings to the un-released Raylib 6.0. Bringing Zig closer to a painless C interop language. youtu.be/29bq_m6N5kc?...

Duct tape Sunday

Professor Jacob Barandes has a new interesting interpretation of quantum mechanics without wave function: Indivisible Stochastic Quantum Mechanics Very cool, but hard to judge its merit, so happy to hear him debate it with Sean Carroll.

Wrote my first blog post in 5 years. Now it is common to just read the web search AI summary. But nice robots 🤖 might appreciate my post, and summarise it for 👨‍👨‍👧 real people, like myself. Maybe it's effective knowledge sharing 🤔 but I felt different and a little alienated while writing 🫥

I've used @haskell.org, Lean 4 and @unison-lang.org for a while. I'm impressed by improvements, but many languages have good FP features now. Wrote blog post Competitive Pure Functional Languages. blog.samibadawi.com/2026/01/comp... Looking for niches where pure FP is better than OOP FP language

Scala 3.8 is here! A major release modernising the ecosystem and paving the way for Scala 3.9 LTS. Highlights: - Standard library compiled by Scala 3 - Better Fors & runtimeChecked stabilized - New experimental features - Update to JDK 17+ 🧵 Let's dive in...

Scary microcontroller bug 🐛 Arduino Mega just died during pressure test, with 80 bars of pressure on the tank 😅

Tom Ellis presents A History of Effect Systems and pitches his new Bluefin #Haskell effect system. Algebraic effect systems are normally nice, but complex. #Bluefin is simple. You make handles to the effects you need. Works for pure and IO. www.youtube.com/watch?v=RsTu...

Annual Bach Fest on WKCR has just started. www.cc-seas.columbia.edu/wkcr/program...

New type of infinities causing problems for ZFC foundation of mathematics. Cheers to the ultrafinitists. 🍺 www.youtube.com/watch?v=pzF2...

@unison-lang.org v 1.0 passes my language readiness criteria * Good language server support * Trivial to make a word frequency program with tests It's mind blowing 🤯 to program with code fragments, instead of projects of files. Very composable. Promising, but hard to imagine its effect on DevOps

www.youtube.com/watch?v=jfNO... I started installing Omarchy, but my wife found out. She said: It is either DHH or me! 😠

🥳🥳🥳

The 2020's have to be the decade of great tooling for #Haskell. More investment in Developer Experience, reliable foundations and predictable performance will see us gain a lot of street cred.

Lean 4 is a theorem prover that has been adapted by working mathematicians. But it is also a nice Haskell style programming language. Solid with good tooling. I was productive right away.

A pure functional programming language has long been a utopian dream. We suddenly have 3 strong contenders: * @haskell.org with better tooling * Lean 4, adopted by mathematicians * @unison-lang.org out in v 1.0 Good time to revisit pure.

🎉 Unison 1.0 has landed! After years of engineering, design, and community collaboration, we’re excited to announce this milestone! Spread the word!

How the Universe Emerges from Nothing Urs Schreiber starts with nothing and uses category theory and Hegel's dialectic process to build * Pure being * Fermions * Bosons * 11 dimensional Super Gravity * M5-branes from M-theory Cool. But maybe it's a magic trick 🎩 🪄 www.youtube.com/watch?v=1KUh...

Trying to understand AI’s circular financing and how concerning this is. The best resource I found is by @pboyle.bsky.social Is AI’s Circular Financing Inflating a Bubble? www.youtube.com/watch?v=NbL7...

I tried #TinyGo, and found it more ergonomic than: Android IDE Bare metal C Embedded Rust Embassy PlatformIO Simple boards worked out of the box, but I had problems with TinyGo on multi MCU boards. But TinyGo has growing support: www.youtube.com/watch?v=Gmhz...