Thank you for letting us share your thought process on reading a paper. It provides a lot of interesting concrete and less concrete pieces of evidence. For example, the two OOM difference between reading and writing down the thinking about it seems a very useful heuristic and very related to things like scaling laws and the difficulty to communicate hard to train skills.
I’m not sure how representative your process and it’s quality is, but that only means more people should try reproducing it—maybe also only covering up to the introduction—in some way that avoids issues whit judgements later.
For example, the two OOM difference between reading and writing down the thinking about it seems a very useful heuristic and very related to things like scaling laws and the difficulty to communicate hard to train skills.
A lot of this is because my thoughts aren’t in whole sentences, let alone ones that can be understood by other people. So the largest cost multiplier was trying to find the right words such that other people can understand me, rather than just writing down my thoughts.
Which is sort of what I am pointing at here. Some people think we lack smarts (aka think-oomph/ability to form and hold complex representations in mind) to solve complex problems such as alignment or inadequate institutions, but I think it is at least plausible that there is enough smarts and the main problem is sharing and combining of solutions people can already hold in their heads. If thoughts can only be rendered 2 OOM slower than produced (and that seems to be a lower bound given that you didn’t even get to the main part of the paper) and are mostly exchanged with high fidelity between two persons (wider communication often needs to leave out large parts of what a researcher actually knows or holds in their mind). That is a massive bottleneck on percolating solutions thru society.
Thank you for letting us share your thought process on reading a paper. It provides a lot of interesting concrete and less concrete pieces of evidence. For example, the two OOM difference between reading and writing down the thinking about it seems a very useful heuristic and very related to things like scaling laws and the difficulty to communicate hard to train skills.
I’m not sure how representative your process and it’s quality is, but that only means more people should try reproducing it—maybe also only covering up to the introduction—in some way that avoids issues whit judgements later.
Thanks!
A lot of this is because my thoughts aren’t in whole sentences, let alone ones that can be understood by other people. So the largest cost multiplier was trying to find the right words such that other people can understand me, rather than just writing down my thoughts.
Which is sort of what I am pointing at here. Some people think we lack smarts (aka think-oomph/ability to form and hold complex representations in mind) to solve complex problems such as alignment or inadequate institutions, but I think it is at least plausible that there is enough smarts and the main problem is sharing and combining of solutions people can already hold in their heads. If thoughts can only be rendered 2 OOM slower than produced (and that seems to be a lower bound given that you didn’t even get to the main part of the paper) and are mostly exchanged with high fidelity between two persons (wider communication often needs to leave out large parts of what a researcher actually knows or holds in their mind). That is a massive bottleneck on percolating solutions thru society.