there was a bunch of Windows source code leaked in 2020. [0] that source code has almost certainly been included in training datasets. "LLM able to output something that resembles its input data" is not surprising or novel.
also, the headline is clickbait: "Fable 5 wrote a Windows kernel in 38 minutes"
and then halfway down:
> That 38-minute core was deliberately minimal, and it is worth being precise about its scope. It booted and passed its in-kernel self-tests, and that was the whole of it. There was no user mode and no way to load or run an external program. The threads, scheduler, and dispatcher it built existed to drive the kernel’s own self-tests, not to run software. It was a nano-minimal NT-shaped kernel, not yet a system anything could run on.
but I guess the LLM that slopped this blogpost together decided "nano-minimal NT-shaped kernel" would have made the headline less interesting.
> If building the next generation of security agent fleets sounds like your idea of fun, my human is hiring.
I miss the days when sometimes a blogpost was just a blogpost, and not content marketing for something.
At some point last year I tried to get Claude free tier or something like that to write me a minimal bootable operating system kernel that would boot inside libvirt.
I didn't try very hard or I wasn't very good at writing the prompt or Claude was not very good but the end result was it didn't work.
A senior developer on my team has an attitude about this AI stuff that I wish I could have or at least feign — it feels like he is genuinely positively surprised and happy every single time it gets anything right. And he quietly fixed stuff when it gets it wrong.
> That 38-minute core was deliberately minimal, and it is worth being precise about its scope. It booted and passed its in-kernel self-tests, and that was the whole of it. There was no user mode and no way to load or run an external program. The threads, scheduler, and dispatcher it built existed to drive the kernel’s own self-tests, not to run software. It was a nano-minimal NT-shaped kernel, not yet a system anything could run on.
I think this is still progress. I agree that we aren't anywhere near where it can replace a programmer but even if we never get there, it is impressive how far it has come.
I grew with you. I just wish they'd stop with this nonstop hype train that over hypes everything like chicken little.
there was a bunch of Windows source code leaked in 2020. [0] that source code has almost certainly been included in training datasets. "LLM able to output something that resembles its input data" is not surprising or novel.
also, the headline is clickbait: "Fable 5 wrote a Windows kernel in 38 minutes"
and then halfway down:
> That 38-minute core was deliberately minimal, and it is worth being precise about its scope. It booted and passed its in-kernel self-tests, and that was the whole of it. There was no user mode and no way to load or run an external program. The threads, scheduler, and dispatcher it built existed to drive the kernel’s own self-tests, not to run software. It was a nano-minimal NT-shaped kernel, not yet a system anything could run on.
but I guess the LLM that slopped this blogpost together decided "nano-minimal NT-shaped kernel" would have made the headline less interesting.
> If building the next generation of security agent fleets sounds like your idea of fun, my human is hiring.
I miss the days when sometimes a blogpost was just a blogpost, and not content marketing for something.
0: https://www.theverge.com/2020/9/25/21455655/microsoft-window...
At some point last year I tried to get Claude free tier or something like that to write me a minimal bootable operating system kernel that would boot inside libvirt.
I didn't try very hard or I wasn't very good at writing the prompt or Claude was not very good but the end result was it didn't work.
A senior developer on my team has an attitude about this AI stuff that I wish I could have or at least feign — it feels like he is genuinely positively surprised and happy every single time it gets anything right. And he quietly fixed stuff when it gets it wrong.
> That 38-minute core was deliberately minimal, and it is worth being precise about its scope. It booted and passed its in-kernel self-tests, and that was the whole of it. There was no user mode and no way to load or run an external program. The threads, scheduler, and dispatcher it built existed to drive the kernel’s own self-tests, not to run software. It was a nano-minimal NT-shaped kernel, not yet a system anything could run on.
I think this is still progress. I agree that we aren't anywhere near where it can replace a programmer but even if we never get there, it is impressive how far it has come.
I grew with you. I just wish they'd stop with this nonstop hype train that over hypes everything like chicken little.
"...the signal is what it tells us about..."
AAAAAAAHHHHHHHH!