Just priming an immediate availability response is likely going to get poor results. On the other hand, this does bring up an important point, which is that fe

Just priming an immediate availability response is likely going to get poor results. On the other hand, this does bring up an important point, which is that few people have been systemically trying to figure out how to get it to reason through problems. For instance, if you try the pure completion on WiC you get 50% chance (like in the paper) but if you improve the prompt to self-context stuff you raise it to almost 70% (http://gptprompts.wikidot.com/linguistics:word-in-context http://gptprompts.wikidot.com/linguistics:word-in-context).