Igor Zhirkov

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

[Igor Zhirkov @rub_duck_type]

@preshing IMO we are not there and won't get there any time soon, and I believe not with LLMs. But when we get there will we notice?

[Igor Zhirkov @rub_duck_type]

@preshing The danger I would agree with is AGI passing the complexitty threshold of complete illegibility. Humans then wont be able to understand/validate the changes either. A clever AGI may also deceive people into accepting innocent patches with double meaning.

[Igor Zhirkov @rub_duck_type]

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.
https://t.co/HgsOgrmC3W

[Igor Zhirkov @rub_duck_type]

@preshing Giambattista Vico is the philosopher of makers. "The truth is the made" -- nature is explorable (experiments), not knowable; to know a man made thing is to understand its history of design decisions. LLM has no inherent intentional decision making that a human can reconstruct.

[Jeff Preshing @preshing]

From now on, nobody should be allowed to say anything about AI unless they've watched @badlogicgames's talk: https://t.co/7f2gMcRpyT

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.)

[Igor Zhirkov @rub_duck_type]

@ylecun @Ph_Aghion @erikbryn The industry is in a weird spot, so is IT education. Some people higher up in the hierarchy are overly skeptical and others downright delusional. I feel sorry for junior devs who are now trying to educate themselves, fight temptations and misconceptions, and get their first job.

[Igor Zhirkov @rub_duck_type]

@dari0x @jamonholmgren Hope you'll find balance for your own mental comfort. Your boy will also mimic you to learn behavior patterns enabling him to live a happy life. I used to shame myself for every act of self care, and I needed to fix it as we were expecting a kid.

[Igor Zhirkov @rub_duck_type]

@jamonholmgren Happy parents are necessary to bring up a happy kid, so I personally balance engagement with the kid and taking care of myself and wife. But every kid and family are different. We have a calm one and she likes playing by herself, but we also have almost no help from families.

[Igor Zhirkov @rub_duck_type]

@ratlpolicy There is no such thing as a page of text in this scenario. 100 pages of generic detectives, of Russel, of Spinoza, of Heidegger, of Levinas, Deleuze, Bible? And how prepared are you and how much are you engaging? People spend whole seminars discussing a single page of Hegel.

[Igor Zhirkov @rub_duck_type]

@smokinscientist Also reading 20 pages of Heidegger for a class is different to reading a 300 pages sci-fi novel, on many levels. I don't mean novels are stupid, just that they are engaging with readers in different ways, with different gratification and return-of-effort patterns.

[Igor Zhirkov @rub_duck_type]

@ProfBZZZ A bit surprising. I attend philosophy classes in University of Warsaw as a free listener and most require an average of 10-30 pages reading weekly per class, plus textbooks on top of that. Gen Z students seem to struggle with discipline/scheduling but not with reading itself.

[Igor Zhirkov @rub_duck_type]

@anthonykrose @zksync @gluk64 Thank you for being an inspiration @arose, I know whatever you chose to touch next will turn golden. Working with you was one of the best parts of my ML experience (and several of our former colleagues voiced the same opinion).

[Igor Zhirkov @rub_duck_type]

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.
https://t.co/G6g86KsSxP

[Igor Zhirkov @rub_duck_type]

@paulg Followed up the Leuchtturm hype and got one of their A4 sized notebooks. Sadly with a fountain pen it ghosts too much so I went back to using sketchbooks for note taking. Perhaps they have models with higher gsm? Something like Art Talens seems cheaper with better paper.

[Igor Zhirkov @rub_duck_type]

@GergelyOrosz Great article and over my last years of teaching programming I made sure my students knew about it so that they understand the market better. However, taxes are hitting the highest grossing the hardest too. Kills motivation for some.

[Chris Lattner @clattner_llvm]

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: 👇

[Jeff Preshing @preshing]

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.

[Igor Zhirkov @rub_duck_type]

@JoeBloxsome @johncrickett You can use AI as an imprecise calculator too, but it is not the right tool for the job. Compilers carefully navigate a "chaotic" space of target programs where any tiny error in translation escalates to a binary that acts VERY differently from what's intended.

[Igor Zhirkov @rub_duck_type]

@cryptoverse420 @grok @MasihEther @stats_feed Most schools in Russia are shit for ALL subjects. I went to one of the best schools in Petersburg (2nd largest city), 5 good teachers over 10 years. There are maybe 10 schools good for STEM in the country, there kids are screened and trained for competitions in small groups.

[Igor Zhirkov @rub_duck_type]

@enricozb @VictorTaelin Bugs can be in implementation alone and not in spec. Security/robustness stuff especially.