Home | Structure | Resources | Publications | Seminars | Architecture | Join | Links
Members can bring serious machine learning projects from any direction. Althea decides what gets built first.
The directions below are founder research directions and internal testbeds for the lab infrastructure. They are not a limit on what members can work on.
This direction develops a meta-reviewer capable of verifying the factual consistency of peer reviews. The system draws on knowledge graphs constructed from the scientific literature, then checks whether reviewer claims are supported, contradicted, or underspecified by the paper and surrounding field.
This direction investigates the sampling of novel scientific ideas from generative models. The goal is to explore under-represented regions of hypothesis space: ideas that are coherent, technically meaningful, and unlikely to be proposed by a typical researcher from the current community.
Public paper: Alien Science: Sampling Coherent but Cognitively Unavailable Research Directions from Idea Atoms.
This direction develops multi-agent deliberative systems for event forecasting, combining peer prediction mechanisms, proper scoring rules, structured deliberation, and language-model agents. The aim is to narrow the gap between amateur forecasters and superforecasters while making forecasting experiments more reproducible.
This direction integrates the lab's components into an autoresearch loop capable of formulating hypotheses, designing experiments, running evaluations, and using outcomes to decide what to try next. This is where Althea becomes the lab manager rather than just a collection of tools.
Across these directions, the lab needs project-scoped Althea agents, event logs, resource accounting, paper artifacts, ownership tracking, and consent flows. The research is diverse, but the operating system is shared.