Whitepaper Reading Club: Beyond Autoregression and Backprop
Beyond Autoregression and Backprop
Welcome back to the Whitepaper Reading Club. This session we're diving into two recent releases that each influence the field of AI. We'll be going through two topics in this session :
JEV (TypeSafe AI): A new model that skips the whole "write one word at a time" approach. Instead of generating text, it just hands back a decision in a single shot, which makes it fast and cheap. The catch is that they haven't published how it actually works, so part of the session will be figuring out what's really under the hood. We have a theory !!
PC-ALM (Seely & Gould): A paper that questions how we train networks in the first place. It drops the usual backprop step and lets each layer learn on its own, while still ending up close to what backprop would have given you.
We'll go through what each one is really doing, where it holds up, and what's just hype.
Paper / source: [DESCRIPgoTION COMING SOON]
Format
Short introduction and context setting to start.
A short walkthrough of the paper and its key ideas.
Questions and discussion grounded in the original source.
Moderator-led discussion with no promotions.
Session Guides:
[DESCRIPTION COMING SOON]
Thank you Zo House
An exclusive hacker house and co-living ecosystem designed specifically for Web3 developers, crypto founders, and digital creators. Spanning premium locations in Koramangala it serves as a 24/7 collaborative workspace and networking hub where members leverage blockchain-linked identities for global access, real-world asset ownership, and high-end tech amenities. The community actively drives the decentralized web ecosystem through continuous hackathons, demo days.
Shout out to the team for helping us run the 2nd Whitepaper Reading! Hope to do more
About Whitepaper Reading Club
We are a community of founders, researchers, and builders across Singapore, Malaysia, San Francisco, Bangkok, New York, Lagos, Taipei, and Hong Kong. We meet in person every month to read, discuss, and pressure-test the latest blockchain papers, protocols, and technical ideas.
Learn more: Website · Summaries · Calendar
We create detailed, easy-to-understand summaries for each paper and have held 80+ sessions since June 2023, covering Account Abstraction, Parallel Chains, EIPs, the Bitcoin ecosystem, and AI x Crypto. We are ecosystem-agnostic, not for profit, and focused on projects with technical, product, or social innovation.