what is this?

an unstructured (and super long) distillation of my various bouts of inspiration, thought, and rabbit holes.

media i’ve enjoyed recently

• evolution as backstop for RL - gwern
• situational awareness - leopold aschenbrenner
• android dreams - anand majmudar
• how i got here - preston thorpe
• simulating civilization on minecraft: rich vs poor - ish
• no solvency, no security - trae stephens

interests

• broadly: emergent intelligence and how systems can adapt and learn to uncertain environments without guidance
• specifically:
  • building computational decision models from neurological first principles
  • adversarial attack modeling (red teaming simulations, anomaly detection, wargames)
  • evolutionary design, cellular automata, conway’s game of life
  • quantum computing
  • sci fi terminals and mechanically minimal interfaces

tunnels

12/14/25

there are 2 types of “talent” in the world:

1. those who define success on finite scales: getting a top engineering offer in sf, earning a PhD for the name brand, building a company to optimize numbers.
2. those who define success on immeasurable scales: scientists trying to build agi, engineers trying to establish the foundations for cross-planetary civilizations, curious students reading 200 pg books on quantum entanglement out of pure interest and not to get the bare minimum knowledge to pass their physics exam. see (12/12)

these are clearly very one-off examples but I think we can generate many more, and its unfortunate that the media substantially glorifies the former. my bet is that curiosity will eventually outperform ambition and raw intelligence every time.

12/12/25

there are simply endless things in the world to learn. the most rewarding learning happens in the spaces where there is no finite answer and an infinite possibility space.

11/20/25

“production is deterrence and the factory is the weapons system” ”… in 2025, the factory itself is the weapon” - trae stephens

it appears that all institutions and industries are suffering the same problem: it’s extraordinarily easy to produce output (ex: munitions, planes, coded web apps), but the orchestration mechanisms that produce this are ill-designed (ex: decentralized factory production or repeating ambiguous prompts to ai tools).

in 2025, we have all awaken from the illusive dreams of working static jobs all our lives where we can function as factory workers that churn polished machine components. now all of us are forced to act as designers and create systems that can not simply product output — but produce it meaningfully. those who can design these orchestration systems will win in the next decade.

this is another reason why simulation systems that can playtest interactions with multiple diverse agents is so powerful. it allows you to test this orchestration layer directly.

11/20/25