Rikk Hill
@rikkhill.bsky.social
Machine learning and Python, scary stories and pretty pictures
Better or worse than sharing a name with that actress typecast as an old crone?
September 3, 2025 at 3:24 AM
Better or worse than sharing a name with that actress typecast as an old crone?
Related: your brain (probably) has no native ability to calculate exponents, so *can't* run a physics sim of a projectile, but it still has some internal representation of a parabola because you can throw and catch objects just fine
August 28, 2025 at 3:33 PM
Related: your brain (probably) has no native ability to calculate exponents, so *can't* run a physics sim of a projectile, but it still has some internal representation of a parabola because you can throw and catch objects just fine
We wouldn't run a physics sim for, say, predicting house prices, but we would run a physics sim for calculating the trajectory of a projectile in known conditions. For some tasks, for a while, physics sims were the only game in town. Now we have a technology for arbitrary functional approximation
August 28, 2025 at 3:11 PM
We wouldn't run a physics sim for, say, predicting house prices, but we would run a physics sim for calculating the trajectory of a projectile in known conditions. For some tasks, for a while, physics sims were the only game in town. Now we have a technology for arbitrary functional approximation
Since we live in a physical universe bound by physical laws, you could ask this question for literally any task. Deep learning techniques handle all sorts of optical tasks without any explicit model of the physics of optics, even CGI tasks which previously relied on such explicit physical models
August 28, 2025 at 3:03 PM
Since we live in a physical universe bound by physical laws, you could ask this question for literally any task. Deep learning techniques handle all sorts of optical tasks without any explicit model of the physics of optics, even CGI tasks which previously relied on such explicit physical models
uv is a very, very good attempt to unify the python environment toolchain. It's rapidly developed a wide fanbase because it's actually very good!
August 12, 2025 at 1:34 PM
uv is a very, very good attempt to unify the python environment toolchain. It's rapidly developed a wide fanbase because it's actually very good!
Go back 130 years to the Bloomsbury Set and late Victorian radicals, you have all sorts of strange-bedfellow beliefs: eugenics and birth control, spiritualism and scientific materialism, feminism and fascism. These ideas have had wildly different trajectories but at one point made sense together
August 11, 2025 at 1:02 PM
Go back 130 years to the Bloomsbury Set and late Victorian radicals, you have all sorts of strange-bedfellow beliefs: eugenics and birth control, spiritualism and scientific materialism, feminism and fascism. These ideas have had wildly different trajectories but at one point made sense together
During Covid and lockdown I started watching the 2005 TV show Bones, which has a supporting character who's a left-wing, free speech-obsessed conspiracy theorist in the post-9/11 vein. It surprised me how much less obvious this was as an archetype just 15 years later
August 11, 2025 at 12:54 PM
During Covid and lockdown I started watching the 2005 TV show Bones, which has a supporting character who's a left-wing, free speech-obsessed conspiracy theorist in the post-9/11 vein. It surprised me how much less obvious this was as an archetype just 15 years later
Part of that bullishness seems to come as a reaction to people going on about stochastic parrots and jumped-up autocomplete. Maybe you shouldn't let *these* people break *your* brain.
August 9, 2025 at 9:13 PM
Part of that bullishness seems to come as a reaction to people going on about stochastic parrots and jumped-up autocomplete. Maybe you shouldn't let *these* people break *your* brain.
I don't mean to be rude when I say this. Nobody has a great grasp on this, and most of your comments are certainly from people who are parroting criticism they don't understand. But your takes on AI are naively bullish in a way they don't have to be.
August 9, 2025 at 9:09 PM
I don't mean to be rude when I say this. Nobody has a great grasp on this, and most of your comments are certainly from people who are parroting criticism they don't understand. But your takes on AI are naively bullish in a way they don't have to be.
You are right to suggest that the capabilities of AI rest on technological and economic questions, rather than political or moral ones. Have you considered you may have a shaky grasp of the technological and economic aspects?
August 9, 2025 at 8:55 PM
You are right to suggest that the capabilities of AI rest on technological and economic questions, rather than political or moral ones. Have you considered you may have a shaky grasp of the technological and economic aspects?
As an industry insider and someone who generally enjoys your output, I honestly don't understand what you think about this topic. It's like you've been backed into a particular rhetorical stance which you reflexively argue from without engaging with anything.
August 8, 2025 at 11:19 PM
As an industry insider and someone who generally enjoys your output, I honestly don't understand what you think about this topic. It's like you've been backed into a particular rhetorical stance which you reflexively argue from without engaging with anything.
i.e. People who are not great at writing claiming it's the bee's knees because it writes better than them, and then people who can recognise its flaws reacting to that
July 30, 2025 at 12:12 PM
i.e. People who are not great at writing claiming it's the bee's knees because it writes better than them, and then people who can recognise its flaws reacting to that
I haven't seen many people *complaining* it's bad at creative writing, but I have seen plenty of reaction to claims it is good at writing from people who correctly observe that it's not.
July 30, 2025 at 12:09 PM
I haven't seen many people *complaining* it's bad at creative writing, but I have seen plenty of reaction to claims it is good at writing from people who correctly observe that it's not.
(there is the modern field of computational neuroscience, which is very cool, but super low-level, and very far away from everyday general application)
July 21, 2025 at 9:07 PM
(there is the modern field of computational neuroscience, which is very cool, but super low-level, and very far away from everyday general application)
Relevant to your thoughts: this gives you four quadrants, one of which is "actually emulating the mechanisms of human thought", and it's worth noting that very little tractable commercially applicable research in modern AI is in this quadrant
July 21, 2025 at 9:05 PM
Relevant to your thoughts: this gives you four quadrants, one of which is "actually emulating the mechanisms of human thought", and it's worth noting that very little tractable commercially applicable research in modern AI is in this quadrant
In particular it has this framing for thought about AI being on two axes. The first axis is about emulating humans vs optimal task performance, and the second axis is roughly about functional approximation vs "actually doing stuff"
July 21, 2025 at 9:02 PM
In particular it has this framing for thought about AI being on two axes. The first axis is about emulating humans vs optimal task performance, and the second axis is roughly about functional approximation vs "actually doing stuff"
Something that might be worth your time if you haven't read it already is the first chapter of Russel and Norvig's Artificial Intelligence: a Modern Approach. The rest of the book is all technical but the first chapter is the "history and culture" chapter
July 21, 2025 at 8:58 PM
Something that might be worth your time if you haven't read it already is the first chapter of Russel and Norvig's Artificial Intelligence: a Modern Approach. The rest of the book is all technical but the first chapter is the "history and culture" chapter
Oh, the one about "weak TDD" as well
June 12, 2025 at 7:01 PM
Oh, the one about "weak TDD" as well
The one about Raku being a language by gremlins for gremlins.
June 12, 2025 at 6:59 PM
The one about Raku being a language by gremlins for gremlins.
I was a prolific online dater and stats nerd in that era and found the whole thing super interesting. What an innocent time we lived in.
June 11, 2025 at 4:33 PM
I was a prolific online dater and stats nerd in that era and found the whole thing super interesting. What an innocent time we lived in.
My gift to you: the now-defunct OKCupid Blog, which tried to quantify these things and just started a bunch of arguments.
web.archive.org/web/20091121...
web.archive.org/web/20110201...
web.archive.org/web/20091121...
web.archive.org/web/20110201...
Your Looks and Your Inbox « OkTrends
web.archive.org
June 11, 2025 at 4:01 PM
My gift to you: the now-defunct OKCupid Blog, which tried to quantify these things and just started a bunch of arguments.
web.archive.org/web/20091121...
web.archive.org/web/20110201...
web.archive.org/web/20091121...
web.archive.org/web/20110201...
I guess survivorship bias could be a case of "c is time"
April 28, 2025 at 9:30 PM
I guess survivorship bias could be a case of "c is time"