1 paper · 1 filter
Jim Maar, Denis Paperno, Callum Stuart McDougall +1
Prior work suggests that language models, while trained on next token prediction, show implicit planning behavior: they may select the next token in preparation to a predicted futu…