Igor Zhirkov (@rub_duck_type) profile picture

Programmer, researcher, writer. Author of 'Low-Level Programming: C, assembly and program execution' Ex-compiler team, @the_matter_labs

Warsaw/Nantes
Joined May 2010
Photo by Igor Zhirkov (@rub_duck_type)Photo by Igor Zhirkov (@rub_duck_type)Photo by Igor Zhirkov (@rub_duck_type)Photo by Igor Zhirkov (@rub_duck_type)Photo by Igor Zhirkov (@rub_duck_type)Photo by Igor Zhirkov (@rub_duck_type)Photo by Igor Zhirkov (@rub_duck_type)
Generating code from formal specifications is a powerful but fundamentally niche approach. I argue that it makes other, often harder, parts of engineering even costlier: requirement discovery, debugging, live migration, deployment. rubber-duck-typing.com/posts…
2
119
My thoughts about how LLM disrupt grading in the software engineering education, and whether it is good or bad. rubber-duck-typing.com/posts…
79
Igor Zhirkov retweeted
A beautiful reading for queuing theory fans. If I get it right, it correlates with the Shortest Job First intuition: long jobs in this context are tail latency which increases the waiting time of every job in the queue.
New blog post! "Lorenz and Little: How Much Does Your Tail Cost?" takes a look at how much tail latency contributes to the concurrency in a system, and how much that contributes to cost. brooker.co.za/blog/2026/07/2…
1
2
10
1,170
My last project at Matter Labs was a multitarget diagnostic system for ZKSync OS. Builds support minimal, allocation-free errors, or a rich error context. The cost dropped by 2–4% in RISC-V cycles. I wrote about my system engineering inspired approaches: rubber-duck-typing.com/posts…
2
117
Reading what AI doesn't say (this topic genuinely matters to me, and it happens to be my birthday today, so if you feel like it, a repost would be a nice gift 🎂) Something shifting in AI architecture that I don't think gets discussed enough through a safety lens. Models are starting to pass information to each other directly through internal representations, hidden states, activations, and computation vectors (still mostly research right now, but the infrastructure will follow). That matters for how we audit reasoning, since most of what we can currently check comes from what shows up in words. Right now, the main tool we have for auditing AI reasoning is reading chain-of-thought. Korbak et al. (2025) spend a whole paper arguing this window is already "fragile" for single-model reasoning. If models start reasoning through vectors passed model-to-model instead of text, understanding that latent layer becomes the only way to keep any visibility into what's actually happening. Thus I think that what seems underexplored is understanding the math of these latent spaces, and it might be the same research direction as learning how to monitor them. Models appear to converge toward similar geometric structure regardless of architecture (Huh et al. 2024). If that's right, probes for safety-relevant features might generalize: deception patterns, goal representations, misalignment signals studied once and applied across models rather than re-derived per architecture. Whether this holds for safety-critical features specifically is open, so is what adversarially robust latent decoding would even look like. Both feel more urgent than the current research investment suggests.
11
38
13
75
11,725
Making is knowing, it makes things intelligible. This is the foundation of epistemology of the Napolitan philosopher Giambattista Vico, and it explains why we lose the knowledge of software faster than it decays technically. rubber-duck-typing.com/posts…
1
107
Igor Zhirkov retweeted
From now on, nobody should be allowed to say anything about AI unless they've watched @badlogicgames's talk: youtube.com/watch?v=RjfbvDXp… This slide in particular is gold. There are good ways to use coding agents and there are bad ways. These are the good ways! Other people on Twitter are gradually reaching the same conclusions, so save yourself some time and just watch Mario's talk. (I don't think this list will fundamentally change with future models, either — even if the benchmarks reach 100%. Disagree? Tell me why.)
3
2
8
1,710
Why Solidity compilation is much harder than it looks: gas is part of program behavior, “correctness” is underspecified and one-size-fits-all, and every extra compiler or target makes the problem worse. rubber-duck-typing.com/posts…
96
Igor Zhirkov retweeted
The Claude C Compiler is the first AI-generated compiler that builds complex C code, built by @AnthropicAI. Reactions ranged from dismissal as "AI nonsense" to "SW is over": both takes miss the point. As a compiler🐉 expert and experienced SW leader, I see a lot to learn: 👇
81
337
68
2,117
413,287
Igor Zhirkov retweeted
I'm convinced that if you're building a large piece of commercial software, the only good way to use AI coding agents is as a power tool, without forsaking your ability to understand the code. On the other hand, if you're building disposable software — for personal use, for a demo, or just experimenting — then abandoning understanding of the code is totally fine. If the project is simple, the AI-generated code will be easy to understand anyway. I see many takes claiming the opposite. People are saying that soon, understanding the code won't even matter. The logic goes: AI is improving all the time, therefore, it will eventually just do everything without human intervention. Some people are already trying to live in this imagined future: Running Ralph loops, getting agents to supervise other agents, cranking out hundreds of commits per day. More power to them, but I don't plan to work that way. Of course AI will keep improving. In the best case, those improvements will lead us towards greater simplicity. They'll help untangle spaghetti code, suggest better ways to organize the project, generate clear, up-to-date documentation. Projects will become easier to understand and onboarding will become easier for both humans and AI alike. This is the only trend that makes sense to me. I can imagine that in 20 years or so, software engineering will no longer be a glamour job. Development tools will be more intuitive. The opportunities to get rich overnight will fade. We'll still need programmers and system maintainers, but not as many. But I think it will take time.
3
5
1
20
1,831
Igor Zhirkov retweeted
We threw 20 major Ethereum projects — Aave, Uniswap, OpenZeppelin, Solady… — at solc and solx 0.1.1. The result? 8–13× faster than previous versions on heavy builds, beats solc by 5–10× in compile time with via-ir — and cuts gas usage. Let’s dig into the benchmarks 🧵
5
10
8
95
59,277
Igor Zhirkov retweeted
solx beta finally fixes “stack-too-deep” — with no semantic changes. In this post: – How the spill-to-memory fix works – Runtime gas benchmarks – Network-wide savings estimates – Our approach to compiler safety 📖 zksync.mirror.xyz/2e9iV3D4gT…
5
21
1,386
Igor Zhirkov retweeted
⚡️ Solidity turns 10—huge congrats to the @solidity_lang team! ⁣ ⁣ To celebrate we’re shipping 𝐬𝐨𝐥𝐱 𝐛𝐞𝐭𝐚 — a new Solidity compiler built on LLVM designed for efficiency that fixes “𝘴𝘵𝘢𝘤𝘬-𝘵𝘰𝘰-𝘥𝘦𝘦𝘱,” trims gas, and keeps original Solidity semantics.⁣ ⁣ 🧵⬇️
8
24
12
99
35,826
Igor Zhirkov retweeted
Ready to prove Ethereum in seconds on a single GPU? Introducing ZKsync Airbender: The world’s fastest open-source RISC-V zkVM ⚡️
382
539
185
2,016
717,208
Replying to @effectfully
big fan of the Unix philosophy, wish Unix embraced the Unix philosophy
10
60
3
1,900
26,634
Hegelian Spirit minus telos and consciousness plus darwinian thinking is quite similar to a special case of meta system transition, as described by Turchin. I think there was an inspiration there.
106
Igor Zhirkov retweeted
EVM Equivalence on @ZKsync was just the beginning. For years, we’ve been quietly building toward a better future for all of @Ethereum. Introducing ⚡️ solx — a powerful new Solidity-to-EVM compiler 🧵
87
197
52
984
425,372
Who verifies the (zk) verifier? Nethermind! The Formal Verification team at @NethermindSec, led by @JulekSU, just made history by formally verifying the honesty of @ZKSync’s on‑chain zk verifier using EasyCrypt. This is the first time formal verification has been applied to a real-world zk verifier. This breakthrough result confirms the integrity of the verifier, while showcasing ZKSync’s commitment to ZKP security. Huge thanks to @dfirsov, @rub_duck_type, @convoluted_code, @the_matter_labs, and the EasyCrypt team (especially Pierre-Yves Strub) for their invaluable collaboration on this project. Read our blog to see how Nethermind’s pioneering formal verification work is helping secure ZKSync. nethermind.io/blog/who-verif…
2
19
11
106
55,934
All sorrows can be borne if you put them into a story or tell a story about them. (I. Dinesen)
129