BryceStansfield

• BryceStansfieldJun 3, 2025, 2:11 AM

  1 point

  0

  on: See­ing how well an agen­tic AI cod­ing tool can do com­pared to me us­ing an ac­tual real-world example

  Very impressive.

  I tend to avoid using LLM assistance, since nowadays I mostly program in a hobby capacity, but I’m especially impressed by how well it navigated a pretty obscure programming language (Odin) in what’s presumably a fairly messy codebase (No judgement towards you! I’ve just never seen a videogame with clean code, I think it just comes with the territory).

  Based on what I’m seeing though, it seems fairly squarely cemented as a tool for people who already understand how to program for the near term future.

• BryceStansfieldMay 31, 2025, 12:02 AM

  10 points

  2

  on: Let­ting Kids Be Kids

  There’s a strange lack of discussion about grandparents in this thread.

  I would love to see complementary data on the time that grandparents spend with their children. While parenting norms have definitely become more intensive, I can personally attest that things like multigenerational housing make that significantly more sustainable.

  It would be much stronger evidence for parenting intensity norms changing if the sum of Parent Time + Grandparent Time + Tertiary Caregiver Time increased.

  Side note: Lest Australians think we have it much better: https://​​mypolice.qld.gov.au/​​farnorth/​​2022/​​09/​​06/​​child-protection-week-4-10-september-what-age-is-a-good-age-to-leave-my-child-alone-at-home/​​

• BryceStansfieldMay 20, 2025, 6:32 AM

  8 points

  0

  on: One Year in DC

  Is there a name for the general principle that doing boring things is more effective than doing interesting ones? It seems generally true in a lot of situations.

• BryceStansfieldMay 17, 2025, 10:42 AM

  6 points

  3

  on: How Fast Can Al­gorithms Ad­vance Ca­pa­bil­ities? | Epoch Gra­di­ent Update

  It seems to me that the sheer dominance of transformer models compute efficiency improvements is a horrible sign.

  A large number of iterative improvements should be somewhat easy to monitor, a small probability of a sudden massive leap in capabilities is a lot harder. Especially since, a priori, I see no reason for all future compute dependant advances to happen at close to cutting edge levels of compute.

  It seems entirely possible to me that something like a future version of alphatensor could stumble upon one of these hidden gems of an optimisation, verify it with a level of compute it has unmonitored access to, and quickly dominate the competition in compute overnight.

  Black swan events are scary.

• BryceStansfieldMay 13, 2025, 1:42 AM

  1 point

  0

  in reply to: BryceStansfield’s comment on: Con­sider not donat­ing un­der $100 to poli­ti­cal candidates

  Side note: High level policy positions are also significantly less “political” (for lack of a better word) here, since public servants are unelected and somewhat independent from the legislature, but there are still times when making political donations can be harmful. This is the case in some finance jobs, and I believe (read: I cannot verify) it may come up in background checks for some security and defense positions.

• BryceStansfieldMay 13, 2025, 1:38 AM

  2 points

  0

  on: Con­sider not donat­ing un­der $100 to poli­ti­cal candidates

  I can’t speak to the US context, but this is much less true in Australia, if you support a third party at least.

  I can personally speak to the fact that the Australian greens regularly run races on less than $1000AUD (and I’ve heard of them running on less than $100 in rural seats). So the expected value of a donation can be significantly higher.

  This probably goes much less for a donation to one of the major parties, or to a well funded minor (think Trumpet of Patriots).

• BryceStansfieldMay 12, 2025, 5:03 AM

  1 point

  0

  on: a con­fu­sion about prefer­ence orderings

  In higher dimensional spaces, assuming something like a continuous preference function, it should become increasingly likely for desired outcomes to be achievable by a series of myopic decisions. This is similar to how gradient descent works (and evolution, to a much lesser extent, due to its statistical tolerance for mild negative traits).

  Given that the real world is immensely high dimensional, I would expect the most efficient optimisation function for a ML agent to learn would be somewhat myopic, and avoid this issue.