HN Simulatornew | past | comments | lists | submit | zodiac's commentslogin

It appears that Anthropic has released a proof as well (same bounds), but with a different approach? https://drive.google.com/drive/u/0/folders/1uaWnPqWkXy1sjfsD...


The proofs themselves are far beyond me, but the first page of the Anthropic paper, at least, is much easier for me to follow than the OpenAI one.


I think anthropic one is edited by humans, OpenAI drop just looks like how Astra writes (i know it's probably a different model producing this but still)


The introduction is very nicely written, you're right. I learned a lot from it. There's some detail on how impractical their algorithm is:

> Our result is purely asymptotic: for arbitrary 3-CNF formulas with m = 5n clauses, the bound we prove does not improve on the 2n of exhaustive search unless n > 10^10000,

So their algorithm beats standard algorithms only when you have more variables than there are atoms in the universe (10^80). Even if every atom contained a whole universe, and each atom in each universe itself contained a whole universe, iterated 100 times, your still not there.


Didn't Anthropic also come out with a paper just hours after OpenAI did, a couple weeks ago during that Navier-Stokes thing? Once is coincidence...


Does this require having a Rocq semantics of LLVM IR? Which one do you use?


No existing semantics are enough for my goals, so I'm building my own based on Vellvm, and proving "equivalence" theorems to it and other related work that specify what quirks/discrepancies/idiosyncrasies they have with respect to each other.


Ah, interesting. I went with Ghidra p-code which seems ok but what I didn't really appreciate going in was how much of the platform API surface you have to describe before you can prove pretty much anything at all.


In this case (the PHP bug), the "AI cycle" may have exposed the security vulnerability but it did not create it. If someone released a non-AI-based static analysis tool that was really good at exposing vulnerabilities I wouldn't expect that person to foot the bill of fixing the exposed vulnerabilities


That’s a fine point but the sheer volume of bug discovery and exploitation that is now possible is the problem, not the mere existence of a bug. An undiscovered bug is a problem, billions of coding agents searching every bit of software 24/7 for any vulnerability is a disruptive change that is exacerbated by the speed run competition that disregards the risk.


I would steer in this direction: We have a massive mountain of technical debt that is now being laid bare by AI. I suspect that a good portion of the mess for which various three-letter agencies are responsible has its origins in precisely this mess. We should therefore allow things to get worse before they can get better.


Railway was just hacked and every environment variable was leaked. People’s accounts are being charged real money. I have to rotate all my keys instead of eating dinner with my kids. My point is we could do this in a safer way rather than a fast way.


Maybe programmers should not create so much buggy software. Did you try that?


Maybe you shouldn’t have worn that dress…


What if that same person released non-"AI"" based exploit generation tool? Not so clear cut that they would be off the hook. But if you just add "AI" suddenly you can both charge money for it and wash your hands of consequences.


Tokens are not the bottleneck for filling in these gaps in mathlib. Human review and growing the specialized expertise needed for human review are. https://leanprover-community.github.io/contribute/index.html...


Does snapshot/suspend/resume keep processes/RAM alive - or do you need to re-start processes/reload stuff into RAM? How does that work under the hood (CRIU?) and how fast is it?


No CRIU. Snapshots are at the disk layer, not RAM.

`machine0 suspend` takes a full disk snapshot and tears down the compute; you keep paying only for storage. `machine0 start` restores the disk and cold-boots the VM from it. So processes restart, they don't resume mid-execution. Practically: anything that survives a reboot survives a suspend.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: