Each update is dated to the work it describes. For measurements and current availability, follow the links to the relevant page.
Research workflow
A research cycle that keeps the result
Our research workflow can take a proposed experiment, run it within a defined scope and record both positive and negative results.
One recent cycle tested different ways of grouping recorded model data ranges. The work produced a repeatable path from a hypothesis to an experiment and a retained result. The measurements describe that planning task; they do not establish a full-model speedup.
We are using iterative research workflows to propose, test and improve parts of the software that supports our experiments.
People set the scope and checks. AI-assisted proposals are evaluated against those checks, and useful changes can be retained while failed or inconclusive attempts stay in the record. This recursive software improvement is operating under supervision; it is not a claim that the whole research programme runs autonomously.