What is more frightening about this than safe C assembly is that this level of implementation is achievable not with a SOTA model, but with a cost effective model like KIMI. There was human judgment involved in the middle, but reading the article, My reading of the process is as follows:
1.A developer identified the necessity of inline assembly.
2.Defined the safety boundaries for 'memory-safe' inline assembly.
3.Established strict policies for memory access.
4.Curated an allowlist of permissible instructions.
5.Set rigorous test criteria and 'done' conditions.
In short, with the overall guardrails in place, a sub agent loop was run, and this level of code was produced. This raises a number of interesting points about how we should use AI. I haven't looked at all the code, but the idea of passing assembly through safe zones without memory access, and using that as a foundation to achieve this level of implementation through AI, is quite impressive
I find it charming that to distinguish Fil-C from the K&R language they use the term 'Yolo-C'. I have never used inline asm before, I actually didn't realise how much behaviour it's specifying! (When I've needed asm it was on non-gcc compilers)
Edit to add: If I'm understanding this correctly we should be able to run this against projects and detect asm violations, I feel like this would be very valuable to be able to feed these back to maintainers
I wonder if an adversarial user could bypass the checks and achieve memory corruption / code execution. Maybe not a practical attack in most situations but a fun exercise.
> This includes things like asm volatile("" : : : "memory"), which is an old-school way of saying atomic_signal_fence(memory_order_seq_cst).
Not quite. AIUI, the first is just a barrier for the compiler, while the second is also a CPU memory barrier. Godbolt seems to confirm that.
Your godbolt code used atomic_thread_fence
The quote uses atomic_signal_fence.
If you find a way to bypass my checks, file a bug. I tried very hard to break it. My agent loops tried even harder
> My agent loops tried even harder
What happens if you ask to find the strings that will erroneously return True from validateSafeInlineAsm for disallowed asm? :)
It’s surprisingly hard to define „erroneously”, but that’s not far off from what I did
Example of a bug found most recently was that sahf was allowed without a cc constraint.
Anyway, if you find bugs, file them. Would be fun to see if there’s a case me and my agents missed
Oops, you're right. I was thinking of those as nearly interchangeable but they're actually pretty different.
I might give it a try when I have a chance, I'll let you know if anything comes of it.
Unrelated question but since you're here: what's the state of support for Boost?
I was able to build it and a lot of it seemed to work
There was some debugging thing where it embeds debug info using module level assembly that you have to disable.
Thanks! If you don't mind a suggestion: I would probably list it in [1], given how many programs depend on it.
It’s not inline assembly
It’s module assembly
They’re different
I'm confused; did you reply on the wrong thread, or am I misunderstanding something? I was merely suggesting it'd be good to mention Boost on your website and how well it works with Fil-C. This suggestion was not specifically regarding anything called assembly; it was just a general comment. I merely mentioned it because you were here.
Do we have a sense for how many of the programs that work [0] are now detected as having asm violations?
Zero, since I made those programs work back when all inline asm was an error.
So currently most of those still have the hacks to go down the no-inlineasm path when building with Fil-C
For the few where I reinstated the inline assembly, there were no bugs found.
It would be a good experiment to try to reinstate the inlineasm paths in all of the programs that had them. I suspect there’s a low chance of finding a bug if it’s in inline assembly that’s on the critical path.
Interesting, I was wondering where catching these errors would fall between 'silently wrong on certain inputs' to 'how did this ever work!?'
While we're at it, does anyone else want something like a good LSP but for assembly?
I mean one that infers as much context as possible and tries to help as much as possible.
This has to be assembler specific of course. For example, I use fasm which has higher level macros. An LSP could suggest struct fields and other stuff.
The utility margin of SOTA models is greatly overstated. You have to pick seriously niche puzzles to make the money worth spending.
Anyway, this is also very useful for humans to use, so it's mostly a lovely coincidence this level of safety arrived with useful chatbots.
I honestly would be surprised if someone used AI in any other way to achieve this
Same here.