I think the more interesting thing is was it unable to do it even knowing the solution? I feel like only testing a single innovation is biasing the current state of automated AI R&D.