enbao.me

the ocean and the point of it all

reflections from a trip to china

reflections on childhood and being at the beginning

the little things

a healthy man wants a thousand things, and a sick man just wants one

reflections after a week in the bay

detecting adversarial fine-tuning, litmus values, impossiblebench

see as much as you can, as early as you can

more work on agents, bridging the non-verifiable gap, sft vs rl, objective function variants, scheming, lora again

on being shaped by the writers you love

prompt evolution, world models, moe efficiency, synthetic data pipeline

a writing model, rubrics, data filtering, hallucination detection, entropy-based confidence

glm 4.5, persona vectors, and system cards

kimi k2, reasoning length, system prompts, dual-token rlvr

looking into and testing sgd, momentum, rmsprop, adam, adamw, and muon

toxic data, alignment faking, dynamic chunking, entropy-based exploration, general reward models, magistral

seed1.5, generalist reward models, spiral, natural thoughts, 2-simplical attention

autoregressive u-nets, diffusion llms, test-time compute, rlvr, and mechanistic interpretability