hi, i'm james.

right now i'm still figuring out where i fit, mostly by probing model architectures, pre- and post-training, and ai safety. in my free time, i study cs at union.

most recently, i interned at voiceos in san francisco, where i worked on its agent harness. i also developed ai agents part time for pivot point and software for aivres .

here are some of my beliefs:

• don’t settle into a local minimum

  a point where small changes no longer improve things, even though a different path could lead somewhere better.

• agency compounds
• time is scary
• most things are learnable
• a pity system exists

  only if the p is improving between tries.

here are some things i like:

• grokbots, muse, intuition, my douyin algorithm, eating, running, making contents, using llms, piano, saipan, x (twitter), light theme, dictation, ai notetakers, "boson" the word.

questions i keep coming back to:

• when do you keep going, and when do you switch?

if you're drawn to the same things, say hello.