Been learning about ZK-SNARKs lately! The core ideas behind modern hash-based schemes are surprisingly accessible & elegant 🧙‍♂️ Partly to teach myself, I wrote a post explaining them from scratch for people like me without a deep math background: xn--2-kmb.com/zk-snarks

There is now! github.com/transmissions11/z…

thank you!! and yup i got lost in the groth16 sauce last time i tried to see what was up in ZK land heh

yup lots of overlap with thaler's book (ofc i drop a lot of the formal rigor and use some intentionally super basic notation), but more focused on being an end to end walkthru of one particular class of proof systems (including some newer stuff like LogUp, VEIL, etc)

glad you're enjoying it!! yup i'm def being a bit fast and loose with the word proof, tho fwiw "proof" being used to mean "false claim passes w/ very low probability" is pretty standard when writing about ZKPs afaict, but it prob wouldn't hurt to be explicit. i'll add a sidenote

ah sweet, wasn't meant to be a gotcha! iirc hevm wasn't able to prove equivalence between contracts w/ reasonable performance for things with nontrivial loops and such, forge is much better i assume?

how will you verify the changes are safe / don’t affect semantics?

overprovisioned VMs

we've still ~36 orders of magnitude below the theoretical limit on compute per kg! arxiv.org/abs/quant-ph/99080…

unless ofc country of geniuses in a datacenter can speed up compute growth/efficiency at a sufficient rate

this is not necessarily as stupid as it sounds? people genuinely do wonder how profitable inference excluding training spend is for the labs — given we may see forced training slowdowns, scaling of inference demand, etc

Should we book this as a win for controls or actually bad because this is papering over deeper issues etc?

Mmm I think this pretty compelling yea!

hehe

I don’t agree, if you spend time with people at the labs you will realize they’re very earnest and concerned

This anecdote is really interesting, but I don't think "the models don't think they'll succeed in deployment" fully answers things? There's all sorts of ways models are being deployed right now with very little oversight, and in a lot of them the model knows this (people telling their models I'm going to sleep so don't ask me questions, etc)

I don't really follow your former point, but wrt to your latter, Anthropic had misalignment incidents with last generation models in eval settings without fancy/expensive multiagent setups? anthropic.com/news/alignment…

I feel like it's still underdiscussed that seemingly all of the world's ~severe misalignment incidents as of now occurred in evaluation/RL? This is not what LW/MIRI/etc were expecting iiuc. Why aren't we seeing incidents in deployment? Is eval awareness that strong?