Proof of Track 2023-04-21

🔍 ZK

zkml-twitter

With zkML you can guarantee that the model runs correctly and does not reveal the weights, and you can verify that your timeline has not been tampered with through proofs. blog post.

zkFHE

The combination of FHE, which does the computation on the encrypted data, and ZK, which does the proof and trusted computation on the data, ensures both the trusted computation and the computation of the encrypted data in the public environment. code.

ZKML research initiatives

Cathie So wrote some research initiatives in zkML, including exploration of zk weightless models (without float, more zk-friendly), Nova and zkML exploration, etc.

Nova + SuperNova

According to Dai Wei's understanding, Folding is not purely SNARK in nature, and other primitives are added, slides.

🔍 Ethereum

zkvm for edge computing in Ethereum

The advent of Nova has renewed excitement about zkVM, so people're considering zkWASM for edge computing in Ethereum. But it seems to me that it might be essentially a decentralized zkWASM rollup.

Frequency of precompile calls on Ethereum

The most frequent precompile is 124567.

🔍 Layer2

magi

a16z OPStack Rollup client written in Rust. One of the criticisms of Rollup is that it doesn't have enough client diversity (although Ethereum faces this problem too), but I believe that with OPStack, client diversity will get better. A day later, op-erigon is here too, blog...

Thoughts on whether Ethereum fees will be cheaper or more expensive than L1 Solana

It mostly depends on the difference between data throughput and computation throughput. My view is that, for the average user with the average transaction, the gas is low enough that it doesn't matter, unless it's free.

Security of Ethereum Rollup and Celestia Rollup

My conclusion is that Ethereum Rollup is fundamentally more secure, but if we're talking about verification costs, Celestia Rollup is lower.

🔍 MEV

Is random tx sorting useful

In short this design encourages spam, because the MEV bot will try to capture profits statistically.

🔍 Web

How to train your own Large Language Models

This is a guide to some basic LLM training and development techniques. It is important to master these basic skills, after all, once you understand the process, you can let LLM do the details for you.

deepmind and google brain

Merged.

WebGPT

The difference between web-llm and WebGPT is that WebGPT is pure js, both use WebGPU.

Intl.Segmenter

It turns out that browsers have a native word splitter.

firefox automatically handles cookies banners

Funny thing is, I'm using arc, and I got cookies when I opened this page.

🔍 Thanks for reading!

I'm not a financial advisor. Proof Of Track is for education and information purposes only. dYOR.

Subscribe Proof Of Track, follow me at @msfew_eth, or connect with me at sny.is.