kevinventullo

(See also the famous Pascal quote “This would have been a shorter letter if I had the time”).

P.s. for reference I’ve asked an LLM to compress what I wrote above. Here is the output:

When I have a murky idea that’s hard to articulate, I find it helpful to ramble—typing or dictating a stream of semi-structured thoughts—and then ask an LLM to summarize. It often captures what I mean, but more clearly and effectively.

E.g. I still kind of write python as if I was writing C++. So, sometimes I’ll write a for loop iterating over integer indexes and tell the LLM “Hey can you rewrite this more pythonically” and I get decent results.

So when there’s broad agreement that something is a good idea like, I don’t know, making sure the water is drinkable, your reaction to that is it’s “evil incarnate”?

The basic idea is to maintain a buffer of size 2k, run mutable unsorted top k on that, drop the smaller half (i.e the lowest k elements), then stream in the next k elements from the main list. Each iteration takes O(k), but you’re processing k elements at a time, so overall runtime is O(n).

When you’re done, you can of course sort for an additional k*log(k) cost.

I’m not normally this pedantic, but on the topic of Star Wars it somehow feels appropriate.

Many natural questions would arise, starting with “Is this the only counterexample?”

This is obviously false. It would disincentivize the police from collecting evidence without first ensuring that the method of collection was legal.

I don't want to make any strong claims here, but my gut reaction to your first comment is that what one manager calls “allowance for illegibility”, another might call “trust in my reports”.

The basic idea is that the only viable way to administer a complex and heterogenous system like a massive corporation is to simplify by enforcing “legibility” or homogeneity. Without this, central control becomes far too complex to manage. Thus, the simplification becomes a mandate, even at the cost of great inefficiencies.

What makes the book particularly interesting is the many different historical examples of this phenomenon, across a wide array of human endeavors.

So then, in a world where the outcomes of all experiments could be reasonably fed into an AI model, it seems that there could be a great deal of value in having scientists publish these “low value” negative results, even in a relatively informal format (i.e. not worrying about formatting, perhaps skipping peer review, etc.). That way even if a human never reads the paper, at the very least an AI “scientist” model would pick it up.

I don’t think you can professionally incentivize this. Rather, scientists would need to do this out of the desire to contribute to the sum total of human knowledge, which could be embodied and not forgotten by these scientist AI’s.