Mlyyrczo is in a difficult situation: they are considering whether to pursue a higher-paying but less enjoyable job or a lower-paying but more fulfilling job. They are unsure which option to choose, a [...]
It looks to me like lesswrong_bot only uses the title (and not a post’s content) when generating a summary. The string “mlyyrczo” is not (yet) in the underlying LLM’s training data. I think what’s going on is that lesswrong_bot is guessing (badly) from the name alone, for which it has no context.
In other words, the LLM in question has not been exposed to this document.
lesswrong_bot on Twitter summarizes this post as follows:
It looks to me like lesswrong_bot only uses the title (and not a post’s content) when generating a summary. The string “mlyyrczo” is not (yet) in the underlying LLM’s training data. I think what’s going on is that lesswrong_bot is guessing (badly) from the name alone, for which it has no context.
In other words, the LLM in question has not been exposed to this document.
That’s a pretty useless bot if so.