@henrycooke As for the FSM, one could just give this to an LLM as it’s a single workbook. Even then, a single instance needn’t attend to the entire workbook in every turn. But it’s a very different object from the IDI
@henrycooke As you suggest, token cost/context size may not be the main constraint since a single instance wouldn’t need to attend to the whole IDI, for eg, at once—it could use tools to surface the relevant slice
@henrycooke I take your point about Copliot, but when you say “truly next-level governance stuff unlocked by AI,” what capabilities do you have in mind?
@emollick Do you think this shows up as firms outsourcing less overall, or mainly outsourcing different things? While AI makes in-house prod cheaper, external vendors also sell independence, credibility & political cover in which they retain comp. advantage, right?
@BrianCAlbrecht Under what conditions does the analogy hold & what evidence would update you? Seems like it requires AI to be geographically constrained, face diseconomies of scale at some margin, and "human-made" preferences to be as durable as "local" preferences. Are those likely?
@Kleisthenes2 Have you found anything more on this? I thought the Employment Contracts Act (1991) led to an increase in employment among part-time and casual workers--is this just a case that's against the average or is it the result of a different change in employment law?
@Kleisthenes2 Joseph Henrich recounts similar instances of encoded knowledge in ‘The Secret of Our Success’. Here’s an old scholars stage article summarising Henrich and similar ideas: https://t.co/6KjTOnUC42. Apologies if you’ve already seen.
Great article in The Post today about land value rates. Some research we shared:
- There is $3.8B of vacant or underused land in Wellington
- The median homeowner would get a 4.4% rates cut with LVR
- Owners of property worth <$1M would get a median 10.8% rates cut with LVR