i believe ai is the ultimate compressor for the hardest problems and a superpower for productivity. i'm dedicated to accelerating its progress. decided to launch a `/now` page to put the mountain of notes and voice logs i've been accumulating somewhere that actually exists and is shareable. this is where i'll log what i'm reading, what i'm building, and whatever technical puzzles are currently stuck in my head.
the lens i work through: multi-agent systems and reinforcement learning infrastructure for LLMs. right now that means high compute RL, megakernels, large scale retrieval and memory components. i'm also deep in building hyper agentic science workflows with self improvement capabilities. agents that accumulate knowledge, maintain memory across sessions and get meaningfully better at their job every time they run. not better in a vague sense, but measurably: tighter retrieval, sharper reasoning, less wasted compute.
also tinkering about how AI models could perform complex, difficult, long-horizon agentic tasks in ultra-realistic settings — environments that go far beyond what models can do today, where they learn to navigate ambiguity, handle interruptions, maintain context over extended interactions.
the thread connecting all of it is systems that know what to think about, know when they're wrong, and improve in real time once deployed. that's the problem i keep coming back to. some of this will look completely different in a month, but that's the point. let's see where it goes.