Doesn't imply much to be honest, the thing that's most interesting for me in there is the training cut-off – it's good to provide an accurate date to the model so it can usefully refuse to answer questions about events beyond that point

→ View original post on X — @simonw,