my conspiracy theory is that the great monuments of ancient history were a political instrument to keep people busy and poor. being busy and poor keeps the ruled docile.

Operating model for research with agent: - Create research folder /research/ - Plan topic and where to gather resources - At every step, ask to drop the raw data source on a folder /research/data - At every artifact, save it into research/session/artifacts/ - For every artifact, save the code to reproduce it in research/src/ - Git it The idea is that it needs to be reproducible and should be easy to redo analysis later by having the data sources checked in

After AI, pursuing beauty is all that is left

film makers herzog recommends: terence malick, errol morris, florian henckel von donnersmarck

it's either: - model specialization is the product. resources should go to getting rewards signals from use cases - model generalization is the product. get a boat in the water, let the tide rise it. beta levered thesis

when developing i want to track APM or equivalent throughput

TODO: - connect lifelog with twitter - add ability to post from the UI

the positive cases for ai and robotics is to free human from labor in order to cultivate virtue and reason aristotle's ideal city without the slavery part

the best mission driven companies can be started by doing this: 1. Write down your core values 2. Write down the logical conclusions in the limit, eg. if so happens, what becomes true 3. Frame the conclusion as a goal statement 4. Roll it back into shippable intermediaries

product idea: - make claude code sessions replayable - allow people to share the iteration process on how they built something - in RTS games watching replays used to be the way to get meta about different builds, something similar could be done here

I've been nerd sniped by DAGs. It's my hammer and all I see it's nails.

Everyone should have evals of what they're an expert on and automatically check if models are able to complete them

beautiful symbolism in the Great Seal of the US should buy one: https://www.flagcenter.com/u-s-presidential-with-official-seal-flag-outdoor-flag/?sku=president-01&gad_source=1&gad_campaignid=19666384234&gbraid=0AAAAAD0KOvXtr9q-jzbWRQUOG-iHBjvYD&gclid=CjwKCAiAmePKBhAfEiwAU3Ko3BRKtbqVpYOJjN8JNkXUtQZ0cAZgTbXgmLN1MYhWqkSJ2oBKc0jBwRoCmCEQAvD_BwE

A tool to cover the ai infra buildout

If you can't solve a software problem is likely because you haven't built the right tooling

I think the contrarian move right now is to make good retrieval systems. When everyone is going to generative slop, improve the current media with models focusing on better search and retrieval.

Want to get certified in 2026 https://www.modernsailing.com/article/sailing-courses-clinics

mission driven themes worth pursuing: - autonomy and self determination - offsetting national debt - fighting entropy of existing systems - situational awareness of energy + datacenter infra

Thanksgiving cooking notes: - 12lb bird is perfect size. Better to roast two of this size than one giant one - buy 2 days before to dry in the fridge, improves crispyness on roast - preorder at butcher - other ingredients: tomatoes, yellow onions, garlic, parsley, butter, dry apple cider - 3 hrs @ 180 deg approx, rest as much as cooked - internal temp 165degs

Read "In the blink of an eye" which are the memoirs of a film editor In the 90s, the industry went from mechanical to digital when computer memory got really cheap Over night the industry went from Linear Video Editing to Random Access Editing, fully digital. The implications are major and obvious but also subtle and negative. In an extreme exercise to imagine the future, he thinks of a crewless, individual creator that is able to will scenes from his mind and how would that creative process will fare about. This future has arrived.