Dru's Notes

In 2012 I gave a side project two hours of deep work each morning before my day job. It failed and I dropped the habit with it. METR grades AI models by the longest task they can finish, and a person with chopped-up attention is stuck where GPT-2 was.

Show Notes

In 2012 I gave a side project two hours of deep work each morning before my day job. It failed and I dropped the habit with it. METR grades AI models by the longest task they can finish, and a person with chopped-up attention is stuck where GPT-2 was.

Featuring: Ken Thompson, Llion Jones

Watch on YouTube

Read the full post

Brought to you by:

Trends.vc: Discover new markets and opportunities.

HeadsUp: Know when your competitors move. Before your customers do.

Multi: Turn any content into a YouTube video.

Chapters:

(00:00) Attention is all you need
(00:04) Building Shommi before work in 2012
(00:23) Dropping the deep work habit
(00:37) How METR grades AI models
(01:04) Chopped-up attention
(01:09) Ken Thompson's four weeks and Unix
(01:35) Two and a half hours each morning

Where to find Dru:

I write an essay daily. Get them free
Follow me on X
Watch my video essays on YouTube
Follow Dru's Notes on Apple Podcasts or Spotify

Subscribe to Dru's Notes for a new episode every day.

Creators and Guests

Host
Dru Riley
Founder of Trends.vc
KT
Guest
Ken Thompson
Featured in Dru's Notes: The Power of Focus: What Early AI Models Taught Me.
LJ
Guest
Llion Jones
Featured in Dru's Notes: The Power of Focus: What Early AI Models Taught Me.

What is Dru's Notes?

One mental model each day in a few minutes. Habits, business, building products and thinking clearly. From Dru Riley, founder of Trends.vc.

"Attention is all you need."
Llion Jones

When I had a day job in 2012, I got up early to build Shommi, a tool that watched Twitter for people ready to buy and sent them to businesses.

I gave it two hours of deep work each morning before my job. My mind was fresh and my attention was at its highest.

In about three months I had a working product.

Nobody bought it.

Shommi failed, but the habit behind it was right. My blunder was dropping the habit along with the product.

For years after that I moved from project to project without giving any of them two morning hours of deep work.

METR, a research group, grades each AI model by the longest software task it can finish, measured by how long that task takes a person. GPT-2, one of the early models, could finish tasks that take a person about 2 seconds. Give it anything longer and it failed more often than it got it right.

By May 2026, the best model METR measured, an early version of Claude Mythos, could finish tasks that take a person 16 hours or more.

A person whose attention gets chopped up is stuck where GPT-2 was.

In the summer of 1969, Ken Thompson had a month to himself. His wife took their baby son to California to visit his parents. Thompson, an engineer at Bell Labs, split that time into four weeks and gave each week to one piece of a new operating system.

"Essentially one person for a month, it was just my self," he said later.

Those four weeks became Unix, the system macOS still descends from.

Now each morning I meditate, then do about two and a half hours of deep work. This time I want to keep the habit for good.

What could you accomplish with hours of uninterrupted focus?