(locally) nameless one @autoresearchinc (YC26)
Joined February 2020
- Tweets60
- Following192
- Followers347
- Likes55
Jad Ghalayini retweeted
Formal verification is the only way, fully convinced.
Shoutout to Taelin for seeing it before most of us and building it out.
Someone needs to do a YouTube series “formal verification for agent acceleration”
maybe a hot take but i now consider adversarial reviews mostly a waste of tokens. far better to use those tokens on empirical combined with formal verification. i have updated pstack to use the interrogate skill less, and eventually i think i may remove it altogether
i have found very little value with code review in the abstract, agents tend to just get lost in the nitpicks and yagnis
Jad Ghalayini retweeted
Put a heart ❤️ next to a coding problem, and Claude & GPT write recursive code 100% of the time (0% baseline). They also can't tell you why!
Excited to share “A Heart Means Recursion” from my summer at @GoogleDeepMind, on how VLMs read & communicate with visual symbols. 🧵
Jad Ghalayini retweeted
github.com/hiloopai/silmaril…
Run make demo, read a theorem, try to break it. Issues and PRs welcome.
Jad Ghalayini retweeted
Coming next:
- game semantics, with a game model of PCF
- C semantics (Cerberus) and a C compiler written in Lean
- CakeML's compiler pipeline
- more x86, RISC-V and ELF
- IEEE 754 floats and crypto
- GPU kernels in Triton/Gluon
Jad Ghalayini retweeted
Some other theorems I like in there:
- a Metamath proof checker, proved sound
- PCF: adequacy, why it isn't fully abstract, and how parallel-or fixes it
- System F free theorems
- Reynolds' argument that System F has no set-theoretic model
Jad Ghalayini retweeted
The Wasm tools use Binaryen's names and flags: wasm-opt, wasm-merge, wasm-as and so on. Each comes with theorems about what its transforms preserve.
Jad Ghalayini retweeted
sha256sum: whatever you pipe in, wherever ASLR puts the stack, even if the kernel spuriously refuses syscalls, it writes exactly the right hash, exits 0, leaves the filesystem alone, and always terminates.
It's stated in LPS, @theoremlabs Linux model, ported and extended in Lean to x86-64.
Jad Ghalayini retweeted
What's in this release:
- 7 WebAssembly tools with correctness theorems
- a Linux process model for x86-64 and RISC-V
- HOL, Metamath, ZFC and Peano arithmetic
- the lambda cube and parametricity
- three Cambridge courses on semantics and compilers
Jad Ghalayini retweeted
We just open-sourced the first part of our Lean formalization work: Silmaril.
The headline result is a sha256sum binary proved correct from its ELF bytes down to the syscalls it makes. There's a lot more in there, and a lot more coming 🧵
Jad Ghalayini retweeted
Lean's a very sophisticated system. The big advantage of HOL Light is that it isn't, especially with its *three* axioms and tiny kernel.