Konstantin Dunas

@hallerite

I’m a researcher at Prime Intellect. I work on scaling RL & Multi-Agent Systems

Latest essay ·

On the Nature of the Swarm

or, why scaling inference ends in many agents

Open-source work

  • Multi-agent systems Composable agents and programmable environments for training and evaluation.
  • Algorithms layer An expressive abstraction for defining training algorithms.
  • renderers A Python library that converts messages to tokens and back while preserving sampled tokens across turns.

Previously, I built Ludic, an exploration of the RL abstractions that have informed my subsequent work.

Κενὸς ὁ τάφος· τί οὖν ἡ καρδία σου ἔτι μεστὴ φόβου;✝ "The tomb is empty; why then is your heart still full of fear?"