I've used Claude to fix/reconstruct & build leaked Win2k3 on Linux with original toolchain via Wine. This approach included full gdi sources reconstruction. I just don't know what to do with this, it's kinda difficult to "wash" on this scale

Run it as your daily driver and trust your data to it. /s

The whole code base is a vibe coded rewrite, half a year after Bun was acquired by Anthropic.

I see lots of ground for that claim.

I apologize, may I ask you, do you use Bun? If yes, you probably do monitor the development of this project (I do, it sounds reasonable to track your tools/deps), probably familiar with Jared's coding style, decision making process, architecture nuances, previous choices? Do you have any issues opened/closed in Bun's repo? Were you satisfied with contributors' reaction? Do you feel you can trust devteam behind Bun?

I get it if you're trying to defend your buddy, but at the end of the day it's on software to justify itself to me. Not for me (or parent poster) to justify their refusal.

Once bitten twice shy, y'know. Maybe the first bite wasn't even from bun. If bun can't take this on the chin and come back stronger, maybe bun wasn't a good choice to begin with. I'm sure a future version of bun with a rebuilt reputation will have an easy time getting re-adopted by most projects that needed to play it safe during the transition.

1 million lines of code in 7 days = ~6000 lines of code to be reviewed per hour, 24 hours per day.

or... they just trust that their ai got it right, which to most people is "vibe coding".

the speed of the rewrite and various analyses of the resultant codebase provide ample evidence that it was vibe coded and solid SWE practices were ignored

nobody understands the Bun Rust codebase. I wouldn't risk my business on code understood by no person. who is responsible? who will take accountability?

nobody. into the trash with it.

How can you claim following SWE best practices if couldn't realistically even have read the code?

"Please follow best practices."

You're telling me that isn't good enough? You might need to head off to the VC reeducation camps.

So transcoding doesn't work unless every line of code is read? That's not how transcoding is done in practice.

But the code that does the transcoding has been read.

Somebody needs to have read deterministic code to even have a chance of noticing something being wrong.

This has not happened here.

That's just agreeing with extra steps.

1 million lines of code written and approved, in 9 days proves without reasonable doubt it was vibe coded.

It was not ported by an engineer. It was transpiled by an LLM and no engineer has ever seen those 1mloc.

In 7 days?

What are you afraid of?

I'm afraid "we" tackle (agressively) the wrong problem, also making it's tough for the maintainers, who did nothing wrong (I have a lot of sympathy towards Bun's developers, they got a lot of ugly feedback within the last month). I don't think AI-written code is the problem at all. Human signs off the changeset the same way as it happened before. I don't care if Rust rewrite did happen using pipeline/harness and LLMs, if the maintainer takes responsibility, and in projects like Bun it happens "by default", I think.

I agree with you that AI-written code should not be a problem and tons of open-source projects have AI-written code right now. But do you really believe the way Bun rewrites and merges its code to master is the same as before? The change in rhetoric (from "don't overreact, it's just an experiment" to "merge it anyway"), the never-arrived blog post promised to explain the decision are concerning to me.

I really appreciate the maintainers' effort towards this awesome project. However, I think it is fair to be a little bit less confident with the current state of Bun.

A codebase that no human understands.

@zhovner, would you consider reverse engineering of the blobs as a temporary measure? in 2026 it's very doable and scales

We're currently negotiating with Rockchip and will first try to convince them to open source. This particular binary isn't a problem right now. I'm sure it will be open source sooner or later. All efforts are currently focused on hardware validation, and software is being developed only in areas where hardware verification is required.

Since this is a portable battery-powered device, rather than first pushing to open-source the DDR training blob, which is non-resident in memory after it's done it's job and a fairly small binary (less chance for hiding bugs), I'd say it's more important to get open-sourced the support in the BL31 Trusted Firmware for dynamic voltage and frequency scaling of DRAM, and support for maximum power savings in suspend mode.

Rockchip have not fully open-sourced the DRAM DVFS support in BL31, but it's key to achieving full run-time battery life on portable devices - see https://xnux.eu/log/083.html

And the system suspend implementation that Rockchip did open-source in upstream BL31 lacks some functionality compared to their binary BL31, mostly about powering off as many peripherals as possible to save power.

I'm not saying don't bother opening the DDR training, just that these two things are much more important for a portable battery-powered device.

Gamepads usage doesn't match screen actions/movements through the entire presentation video(s), zero trust at this point.

Edit: I apologize, entirely missed the game itself, partially understand the mismatch now, but still, it's much heavier in the presentation than in real life for some reason

> Agentic management software is all the hype today: What started with Moltbot and OpenClaw now has a lot of competition: ZeroClaw, Hermes, AutoGPT etc.

Moltbot is OpenClaw, AutoGPT was born significantly before. I just couldn’t read after the first paragraph, I’ve lost the trust entirely, whatever/whoever wrote it.

Hermes agent dates back to at least September last year too, pre-dating Moltbot/OpenClow by a couple of months https://github.com/NousResearch/hermes-agent/commit/17608c11...

It’s marketing. They’re selling some change management solution, so obviously they advocate for showing AI agents only changes, rather than the full context.

Doesn’t mean it’s a good idea, though.

This usage reset you did on April 23 will not mitigate the struggle we’ve experienced. I didn’t even notice it yesterday. I checked this morning and it came down from 25% weekly to 7%. What is this? I didn’t have problems for two months like many others (maybe my CC habits helped), but two weeks were very painful. Make a proper apology, guys. This “reset” for many users could hit the first days of the week, tell me you thought about that.

And as a bonus: GPT is slow. I’m doing a lot of RE (IDA Pro + MCP), even when 5.4 gives a little bit better guesses (rarely, but happens) - it takes x2-x4 longer. So, it’s just easier to reiterate with Opus

Yeah, need some good RE benchmarks for the LLMs. :)

RE is very interesting problem. A lot more that SWE can be RE'd. I've found the LLMs are reluctant to assist, though you can workaround.

What is RE in this context?

Reverse engineering

I've been messing with using Claude, Codex, and Kimi even for reverse engineering at https://decomp.dev/ it's a ton of fun. Great because matching bytes is a scoring function that's easy for the models to understand and make progress on.

I want to get into RE with AI. Which model you liking the most?

This. People drastically underestimate how much more useful a lightning fast slightly dumb model is compared to a super smart but mega slow model is. Sure, u may need to bust out the beef now and then. However, the overwhelming majority of work the fast stupid model is a better fit.

Mind sharing the use cases you're using IDA via MCP for?

In this new AI-driven world, ideas mean everything; one more year - it will be battles of ideas (not implementations as before).

Yes, we're all very excited for the many AI-created projects that have been created outside the shovel-selling business.... wait. There are none.

There's a lot of 'single serve' software being written now by AI. People using Claude Code to make stuff that solves problems they have. It's wild watching people who don't know how to code just use it to solve problems they have. Even if the solutions can be considered awkward by traditional software engineering standards, to the people just looking to solve their problems, that doesn't matter, so long as it works. I'm a software engineer by trade and don't know shit about ML, but I want a nice tool to be able to do RLHF / DPO on Z-Image, so I'm working with Claude to build one, and so far it can use ComfyUI to generate the image pairs, and allows you to pick A vs. B then start a training run with layer offloading enabled so it fits in 16GB VRAM, and I haven't finished a training run yet, but steps are increasing and loss is changing so... I dunno... I see lots of software being created that wasn't before.

These are all local, though - if ideas were all that mattered, we'd see widely available ones, too.

I am not seeing them. (I would love to be proven wrong, because "how well does this work for not-one-off software" is a really important question for me)

So basically like Excel since the 80s?

Guess we should have just stopped at Excel then?

Yeah, totally, just one more year.

you can do it on wasmer's workers, their last wasm/python approach is pretty solid (compatibility, performance). it's sad to say, but after 4 years of "beta" Python support on CF workers - it's still ugly. I dunno who was responsible for such a neglect, but even with the last changes - total fiasco

Why is it “ugly”?