Posted by lwhsiao 4 hours ago
Wouldn't that be a good cost cutting measure for some of the agent networks?
Honestly, the model probably saw the test set and then modified weights to achieve the outcome. Many, many researchers these days are not being sufficiently critical of results that arise out of training to the test / leaking information about the test. Writing style remains a useful heuristic. These folks aren't serious researchers.