1/2) writing programs and then proving each of them separately is too much job, instead we can prove correctness of metaprogram which generates other programs, so correctness of metaprogram implies correctness of each program it generates.
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
i am waiting RSI guys who will suggest to reimplement this scripts to use new algorithm in order to decrease kappa recursively faster 🚀 🚀 🚀
Validated and merged Rohan's PR. Big gain on a really difficult regime (that frankly I was stuck at). Incredible work.
κ = 2⁻¹⁵ (tightened from κ = 2⁻¹⁸²)
A 500 thousand fold improvement over the previous result and a 2 ^ 167 fold improvement over the original OpenAI result.
This result was a true community effort that combined techniques from a number of contributors. I want to take time to highlight each of them, but for now getting these results verified and published as fast as possible so everyone working on the problem is at the leading edge.
it is so interesting that we can do 2-4 time better lossless compression of images and text by using LLMs. But speed of compression/decompression is about 500 time slower in CPU
nature.com/articles/s42256-0…
back in 2015, @DARPA spent over $25M(and even more) on Crowd-Sourced Formal Verification program while SAT solvers able to solve that games in less than seconds. wtf is going on in US?
darpa.mil/research/programs/…
rumors say anthropic fixed llm hallucinations by running them at 0 kelvin. this is how FLT gets proved
