His research shows that many of the current claims of emergence appear to be the result of a skewed way of measuring the phenomenon. If this is the case, we could be in a bigger pickle in terms of AI safety and alignment if our detection methods for emergent abilities are flawed.
