Jacob Marshall

multi armed bandit

prev: research @sievedata, co-founded quant trading

[Jacob Marshall @jcbmrshll]

what are some good grants for novel benchmarks? my rollouts are mega expensive

[Jacob Marshall @jcbmrshll]

@radbackwards noooooooooooooo

[Jacob Marshall @jcbmrshll]

@pandoks It’s Barcelona

[Jacob Marshall @jcbmrshll]

San Francisco looks great today

[Jacob Marshall @jcbmrshll]

had been suspecting that surely someone was working on this for RL :) cool launch!

[Exa @ExaAILabs]

We built a time machine for the web.

Introducing Exa Snapshot: an index of 400 billion historical snapshots of webpages that lets you search as if it's the past.

Snapshot is already being used for backtesting prediction models, RL at labs, exploring the pre-AI web, and more.

[Jacob Marshall @jcbmrshll]

In another life I'm professional poker player Jacob "Stacks" Marshall. Nice to still be able to win tournaments in this life though.

[Jacob Marshall @jcbmrshll]

@brendanashwrth nice work!!

[Jacob Marshall @jcbmrshll]

@SkyeSharkie many such cases, incentives can get so misaligned

[Jacob Marshall @jcbmrshll]

@evanjconrad so cool!

[Florian Brand @xeophon]

an unspoken rule among eval creators is to come up with the most cursed infra setups to make everyone suffer

[Jacob Marshall @jcbmrshll]

@fujikanaeda The quality of direction I can provide agents deteriorates as session count increases. There are plenty of aids that could help address this, both from agent as well as just in UX. Also maybe this means I should just have better instructions at prompt 1 but that has its own costs

[Jacob Marshall @jcbmrshll]

fabulous lecture!

[Eric Jang @ericjang11]

For the last few months I've been working on a from-scratch implementation of AlphaGo, a 2016 AI breakthrough that inspired me to get into deep learning. My casual understanding of AlphaGo was "search-augmented deep neural networks trained with self-play", but I wanted to go deeper and understand it by creating it.

Frontier deep learning research has always been expensive, but any given capability gets cheaper very quickly. In 2026, you no longer need DeepMind's resources to train a strong Go AI - you can vibe code all of it yourself for just a few thousand dollars of rented compute.

It was a huge honor to be invited to teach this with @dwarkesh_sp on @dwarkeshpodcast

I am an AlphaGo & Go apprentice, not a master, so all factual errors in the podcast are mine.

Web version of tutorial: https://t.co/Xkf9VsgtuT
Code: https://t.co/rWKOwclPDg
Play the go bot here: https://t.co/aVglJXldVX

[wordgrammer @wordgrammer]

Despite 3 years of progress, AI is still worse than me at stuff I understand, but unfathomably brilliant outside my area of expertise. Wonder why this is

[Justin Skycak @justinskycak]

One of the biggest factors in elite performance is the sheer volume of action-feedback-improvement cycles.

[Jacob Marshall @jcbmrshll]

e.g. they will gleefully leak your val set into a prompt

[Jacob Marshall @jcbmrshll]

there are niche areas where you can tell coding agents have really bad reward hacking

[Jacob Marshall @jcbmrshll]

@jackminong I think watching really good players talk through a hand on YouTube is the quickest and best way to get good intuition

[kache @yacineMTB]

do not think that you can avoid doing this. you need to understand. you can outsource thinking, writing code, but you can't outsource understanding. to know where you can go, you must know where you are

[Jacob Marshall @jcbmrshll]

all these git worktree wrappers are hilarious