Get live statistics and analysis of Joseph Suarez 🐡's profile on X / Twitter

I don't like courses. Most were a waste of time. Yes, even at Stanford. If you're new to ML, take CS231N.

My Advice for Programming and ML This article is a prequel to my opinionated reinforcement learning guide. It exists because people have asked me for it repeatedly. Start here if you a) do not know how to program b) don't know ML or

You have to have been in ML for over a decade to really understand how bad this was. From 2015-2018ish, any time anything went wrong in an experiment, at least someone would say "local minimum"

I don't care if Karpathy is down on RL. He, Carmack, Ilya, and Alec Radford could all show up in person to tell me I'm wasting my time and I'd still keep doing it. Because damn it this tech is too cool not to exist.

The Tragedy of Reinforcement Learning Or: how the genie went back into the bottle for years. This is, at least from my perspective, the true story of what happened to reinforcement learning and why it's only really starting to kick off

An Ultra Opinionated Guide to Reinforcement Learning Reinforcement learning is about learning through interaction. Applications include robotics, logistics, gaming, and even control problems in science like nuclear fusion. It's an underexplored niche of

RL really sucks. It takes 10 hours just to learn breakout. ... a few years ago. It's <30 seconds on 1 GPU now in PufferLib and still dropping. Write faster code.

If you ask me to do this I will 1) not be able to 2) assume you're a moron who tf remembers toy algos from CS101 with 0 application

PufferLib 3.0: We trained reinforcement learning agents on 1 Petabyte / 12,000 years of data with 1 server. Now you can, too! Our latest release includes algorithmic breakthroughs, massively faster training, and 10 new environments. Live demos on our site. Volume on for trailer!

Accidentally said "you all are status obsessed hollywood wannabees" instead of "you're literally building the future bro" and they kicked me out of SF

25 pages of educational material going live today!

Reinforcement Learning on a Petabyte of Data at Home 6.2 petabytes uncompressed. Our agents played >12,000 years of Neural MMO 3 on a single 6x4090 tinybox in 80 hours. You can watch the agent live in your browser at puffer.ai. Everything

Welcome to RL. If the pink curve overtakes the green curve, we'll have big news

Why RL Failed to Bootstrap Off-policy RL is slow. On-policy RL is data hungry. I figured out why. This is a precise technical summary of my findings from the past 3 weeks of research and literature review across various

I don't care if Karpathy is down on RL. He, Carmack, Ilya, and Alec Radford could all show up in person to tell me I'm wasting my time and I'd still keep doing it. Because damn it this tech is too cool not to exist.

Offline RL is not RL. RL is about interaction. No interaction, no RL.

If you ask me to do this I will 1) not be able to 2) assume you're a moron who tf remembers toy algos from CS101 with 0 application

RL really sucks. It takes 10 hours just to learn breakout. ... a few years ago. It's <30 seconds on 1 GPU now in PufferLib and still dropping. Write faster code.

The Tragedy of Reinforcement Learning Or: how the genie went back into the bottle for years. This is, at least from my perspective, the true story of what happened to reinforcement learning and why it's only really starting to kick off

This is one of my favorite books. I've never read it. Created the easiest way ever to spot a snake.

Welcome to RL. If the pink curve overtakes the green curve, we'll have big news

My Advice for Programming and ML This article is a prequel to my opinionated reinforcement learning guide. It exists because people have asked me for it repeatedly. Start here if you a) do not know how to program b) don't know ML or

PufferLib 3.0: We trained reinforcement learning agents on 1 Petabyte / 12,000 years of data with 1 server. Now you can, too! Our latest release includes algorithmic breakthroughs, massively faster training, and 10 new environments. Live demos on our site. Volume on for trailer!

You have to have been in ML for over a decade to really understand how bad this was. From 2015-2018ish, any time anything went wrong in an experiment, at least someone would say "local minimum"

Accidentally said "you all are status obsessed hollywood wannabees" instead of "you're literally building the future bro" and they kicked me out of SF

Listening to this dude is a waste of time

I don't like courses. Most were a waste of time. Yes, even at Stanford. If you're new to ML, take CS231N.