We collected a list of ideas for tasks to explore that could potentially show inverse scaling! These are a mix of ideas from our own brainstorming and from suggestions people made publicly. We have included links to the original suggestions where applicable.
A few people have asked to read my personal research statement for Ph.D. programs in machine learning, so I’ve released my statement. The statement follows the below outline, which I think is a generally good outline for such statements:
Often when we want to answer a question, we aren’t just interested in the answer, but also in why we think that answer is true. In case we’re wrong, we’re also interested in why other answers could be true.