αlβε

@Perpetualmaniac @schrockn @jarredsumner > is better it's not

@__hl1n__ @schrockn @jarredsumner bun’s “slop” rewrite with AI is better than a carefully crafted zig program by ape coders tells me everything i need to know on which language my AI should write in

I have never actively wanted a programming language to fail but thanks to Andrew Kelley's scurrilous, outrageous, classless post on @jarredsumner 's bun rewrite, I got there. Port every project from Zig to Rust.

At TigerBeetle, as systems engineers, we: 1. write code, sharpening skill and second order understanding 2. read code, walking through the building to constantly improve things 3. review code, ensuring no code is merged without two engineers’ understanding and review.

@filpizlo You know how free software advocates always say, "free as in 'free speech', not 'free beer'"? Maybe they are wrong and it's actually "free" as in "use after free".

Rust is the Paper Straw of Computers It solves a problem nobody had and makes everything worse. And those in power demand that you use it.

@digvijoy_c @antirez ok, I'll tell you some of my merged PRs over the last 20years, if you tell me how much did you pay >1k unverified followers.

@albertodeprezzo @antirez Ok one chance at civility : name a software which has 1/1000 usage of Redis that you were involved in. If the answer is No then you are clueless about “Anthropic will replace…”

People for years told me why I never talked about competitors of Redis. The Zig / Bum drama is a good warning. Do your job, without thinking everything is some form of competition or comparison. So: do your stuff, leave people that act lawfully alone.

@antirez ok, you don't care. Good for you, mate. Maybe there's a guy that wants to sort things out, though, because he consider his project something precious deserving to tell the full story, cleaned of all the noise AI market and Rust community generated over the weeks.

@albertodeprezzo You have no idea how little a fuck I could give. Our personal stories are nothing compared to human progress and evolution.

@digvijoy_c @antirez > trolls who never shipped this thread summarizes your inability to hold a civil debate with unknown people guilty of not being popular

@antirez @albertodeprezzo Thank you for dealing with trolls who have never shipped anything .

@_ontologic @oleksandr_now @antirez he said what everybody knows is true: bun has been poorly managed since the day 2, when instead of solving issues jarred introduced workarounds. Now he can sell it as a success story, which is not, unless you consider the Anthropic acquisition as _the_ discriminant.

It does actually matter how you say things, insisting otherwise is cope by people who will never be taken seriously because they aren’t a serious person. If Andrew wanted the content of his argument to matter more than his tone he wouldn’t have written so acerbically. I’m willing to bet that Andrew has gotten exactly the response from ppl that he was aiming for

@oleksandr_now @antirez yes, indeed. So what's wrong in explaining what's happened? Andrew used some harsh words probably he should have avoided, just because people start focusing on them instead of what he said, which is a perfect snapshot of what bad management is, in the AI age

@albertodeprezzo @antirez but this had happened in both directions like dozens if not hundreds of times?

good points. I would add that bad technical management remains bad even with a strict compiler like rustc. It just produces different technical debt, inconsistencies and (more generally) hidden bugs. The point is: why the IT industry is considering a "passing tests" LLM rewrite "modern programming"? nobody cares about the quality, no more? Probably nobody understands the new codebase, which is something crazy, if I had to adopt it in prod

so many vibes to that andrew kelley post .. > We became increasingly horrified at the programming practices we saw in Bun's codebase. Hacks on top of hacks. Abuse of assertions. Most of all, recklessly speeding past feature after feature with very little time taken for reflection and elimination of bugs and technical debt. does this mean Zig is difficult to scale in a team ? Or just bad software engineering practices were followed .. Any experience to share from developers working in Zig in a team of meaningful size ?

🐧 Linus Torvalds gets angry when people say 99% of our code is written by AI, he told the audience at the Open Source Summit.

Announcing uv audit: native support for vulnerability scanning across your project's dependencies

4. true logit-based distillation isn't possible via standard APIs, and that API distillation isn't magic or lossless. However, response distillation with strong CoT data is real, effective, and widely used in 2026 (including by DeepSeek themselves for their open models). It has enabled huge progress in smaller, efficient reasoning models. It's not "sci-fi". it's engineering that works better than expected.

3. Pretraining provides the broad capability surface. SFT / imitation on high-quality CoT (from teacher) provides the specific reasoning patterns. RL (with verifiers or outcome rewards) refines it further. The "info channel" is the synthetic demonstrations, especially long reasoning traces. This has been shown to transfer reasoning effectively (see DeepSeek's own results and follow-ups). No, it's not as rich as full logits every token, but it's sufficient for major gains. Absence of logits doesn't make it "nonsense." It's just a different regime

Another important thing: Chinese models are not strong because they distill US models. Distillation of models via API is *impossible*. If somebody tells you the contrary, they don't understand machine learning:

@valigo "Go doesn't provide assertions. They are undeniably convenient, but [...] programmers use them as a crutch to avoid thinking about proper error handling and reporting." if you complain about it, you're missing what error is in go: a value you have to handle, not an unwanted case

Just learned that Go doesn't have assertions. How is this language even real.

@ThePrimeagen no surprise. It's just the way a portion of sw industry seems to be working, now. If it compiles, ship it, AI will -eventually- fix it later.

I super don't understand this. I have to believe you're writing no software of any consequence for this to be true.