Reposted by atticus goldfinch
part of it i rarely see commented on is how much of this is just shaming mothers for trying to raise their kids as best they can. the endless optimization, the vaccines, the formula, the dietary restrictions, the (not so much) subtext is "if you don't follow this thick ass book you're a bad mom"
December 8, 2025 at 10:46 PM
part of it i rarely see commented on is how much of this is just shaming mothers for trying to raise their kids as best they can. the endless optimization, the vaccines, the formula, the dietary restrictions, the (not so much) subtext is "if you don't follow this thick ass book you're a bad mom"
Reposted by atticus goldfinch
the number of different ways in which this single sentence is profoundly offensive is just remarkable
RFK Jr: "All of the ingenuity of corporate America has not produced an infant formula that is superior in nutrition and all the qualities that we want to the infant formula that God made, which is the infant formula in a mother's breast."
December 8, 2025 at 10:45 PM
the number of different ways in which this single sentence is profoundly offensive is just remarkable
Neat little new mode of success I just encountered under the new agentic paradigm. I couldn't replicate a bug, but was able to fix it by specifying what was allegedly happening, not what I was seeing locally.
Would've been miserably hard in the past without coding agents. New methods are emerging!
Would've been miserably hard in the past without coding agents. New methods are emerging!
December 8, 2025 at 10:44 PM
Neat little new mode of success I just encountered under the new agentic paradigm. I couldn't replicate a bug, but was able to fix it by specifying what was allegedly happening, not what I was seeing locally.
Would've been miserably hard in the past without coding agents. New methods are emerging!
Would've been miserably hard in the past without coding agents. New methods are emerging!
I love these snapshots of old training dates, it's about as close as you can get to proving objectively that something is very very wrong with the country.
The machine built by feeding it decades of data and news stories is like "shut the fuck up, that shit can't be true."
Immune to frog boiling.
The machine built by feeding it decades of data and news stories is like "shut the fuck up, that shit can't be true."
Immune to frog boiling.
December 8, 2025 at 9:37 PM
I love these snapshots of old training dates, it's about as close as you can get to proving objectively that something is very very wrong with the country.
The machine built by feeding it decades of data and news stories is like "shut the fuck up, that shit can't be true."
Immune to frog boiling.
The machine built by feeding it decades of data and news stories is like "shut the fuck up, that shit can't be true."
Immune to frog boiling.
Reposted by atticus goldfinch
i asked chat gpt to fact check an article for me
December 8, 2025 at 9:33 PM
i asked chat gpt to fact check an article for me
Reposted by atticus goldfinch
Me: *Enters in news about the current administration to the AI Continental Congress*
AI Samuel Adams: "I have some thoughts on which method of hanging is best."
AI Samuel Adams: "I have some thoughts on which method of hanging is best."
December 8, 2025 at 8:50 PM
Me: *Enters in news about the current administration to the AI Continental Congress*
AI Samuel Adams: "I have some thoughts on which method of hanging is best."
AI Samuel Adams: "I have some thoughts on which method of hanging is best."
Reposted by atticus goldfinch
IOW, devs have honed priors from education & experience.
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
I'm writing my thoughts on this now, but my current strong belief is that the most valuable new skills to be combined with technical prowess are epistimics and metacognition.
Do you know how you know things? How you know to debug? To convey that to an agent?
It's explicating tacit knowledge.
Do you know how you know things? How you know to debug? To convey that to an agent?
It's explicating tacit knowledge.
December 8, 2025 at 8:58 AM
IOW, devs have honed priors from education & experience.
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
honestly, ai "versions" of the founders made from a ton of historical documents would be really interesting to do this with, because they'd all immediately start calling for impeachment at _minimum_.
Oh good, Glenn Beck has created an AI George Washington. You will be shocked to learn that the AI George Washington created by Glenn Beck sounds exactly like what would happen if Glenn Beck built an AI George Washington to sound exactly like Glenn Beck.
December 8, 2025 at 8:05 PM
honestly, ai "versions" of the founders made from a ton of historical documents would be really interesting to do this with, because they'd all immediately start calling for impeachment at _minimum_.
Reposted by atticus goldfinch
stupid cyberpunk
Oh good, Glenn Beck has created an AI George Washington. You will be shocked to learn that the AI George Washington created by Glenn Beck sounds exactly like what would happen if Glenn Beck built an AI George Washington to sound exactly like Glenn Beck.
December 8, 2025 at 8:01 PM
stupid cyberpunk
Reposted by atticus goldfinch
ahh I love this
x.com/MattLutzPhi/...
x.com/MattLutzPhi/...
December 8, 2025 at 6:47 PM
ahh I love this
x.com/MattLutzPhi/...
x.com/MattLutzPhi/...
Reposted by atticus goldfinch
This doesn't mean that personas can't be useful - for example, they change how the AI answers questions, the format of output, and maybe other factors as well.
However, you aren't going to make the AI suddenly better through role-play.
Paper: papers.ssrn.com/sol3/papers....
However, you aren't going to make the AI suddenly better through role-play.
Paper: papers.ssrn.com/sol3/papers....
Prompting Science Report 4: Playing Pretend: Expert Personas Don't Improve Factual Accuracy
<span>
<p><span>This is the fourth in a series of short reports that help business, education, and policy leaders understand the technical details of working w
papers.ssrn.com
December 8, 2025 at 6:33 PM
This doesn't mean that personas can't be useful - for example, they change how the AI answers questions, the format of output, and maybe other factors as well.
However, you aren't going to make the AI suddenly better through role-play.
Paper: papers.ssrn.com/sol3/papers....
However, you aren't going to make the AI suddenly better through role-play.
Paper: papers.ssrn.com/sol3/papers....
Reposted by atticus goldfinch
We tested one of the most common prompting techniques: giving the AI a persona to make it more accurate
We found that telling the AI "you are a great physicist" doesn't make it significantly more accurate at answering physics questions, nor does "you are a lawyer" make it worse.
We found that telling the AI "you are a great physicist" doesn't make it significantly more accurate at answering physics questions, nor does "you are a lawyer" make it worse.
December 8, 2025 at 6:32 PM
We tested one of the most common prompting techniques: giving the AI a persona to make it more accurate
We found that telling the AI "you are a great physicist" doesn't make it significantly more accurate at answering physics questions, nor does "you are a lawyer" make it worse.
We found that telling the AI "you are a great physicist" doesn't make it significantly more accurate at answering physics questions, nor does "you are a lawyer" make it worse.
Reposted by atticus goldfinch
some significant part of this is just catch-all fear for “this is not the child I ordered” I think
Oh, here's a big surprise. The transphobes don't believe in ADHD either.
then, the transphobes , feeling victorious at how many people they have made unhappy and despairing in the previous week, move on to anything else that has a spectrum and the possibility of more colours than they can conceive - such hateful people - they cannot understand anything that is not THEM
December 8, 2025 at 1:09 PM
some significant part of this is just catch-all fear for “this is not the child I ordered” I think
Reposted by atticus goldfinch
Good on Penn Jillette. Having to come to terms with your entire political identity and philosophy being wrong is not easy and most people could never do it.
December 7, 2025 at 9:17 PM
Good on Penn Jillette. Having to come to terms with your entire political identity and philosophy being wrong is not easy and most people could never do it.
IOW, devs have honed priors from education & experience.
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
I'm writing my thoughts on this now, but my current strong belief is that the most valuable new skills to be combined with technical prowess are epistimics and metacognition.
Do you know how you know things? How you know to debug? To convey that to an agent?
It's explicating tacit knowledge.
Do you know how you know things? How you know to debug? To convey that to an agent?
It's explicating tacit knowledge.
December 8, 2025 at 8:58 AM
IOW, devs have honed priors from education & experience.
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
But can you articulate those priors -in all their n-dimensional hypervolumetric glory- to an llm?
And do you have the metacognition to understand how to communicate it to a human-like-but-distinctly-different llm so it translates correctly?
I almost made an exponent analogy for LLMs yesterday, now I'm angry I didn't.
You don't want to have (0,1) skill developers and hand them these tools. You just can't afford it. Bad devs are now worse than ever while great ones spin gold on four worktrees simultaneously.
You don't want to have (0,1) skill developers and hand them these tools. You just can't afford it. Bad devs are now worse than ever while great ones spin gold on four worktrees simultaneously.
my new-old blog is not as cool as @nonbinary.computer's but I like the bells and whistles I stapled into it and I proved a lot of things about How To Build Software, to myself at least
in this installment we try to figure out how to at least name the phenomenon where people turn into slop conjurers
in this installment we try to figure out how to at least name the phenomenon where people turn into slop conjurers
Cognitive exponents and LLM leverage
I know a few people for whom LLMs have been a near-immediate multiplier of attention and effort. I know a lot for whom LLMs clearly make them worse at thinking and doing things. So: why?
ed3d.net
December 8, 2025 at 8:21 AM
I almost made an exponent analogy for LLMs yesterday, now I'm angry I didn't.
You don't want to have (0,1) skill developers and hand them these tools. You just can't afford it. Bad devs are now worse than ever while great ones spin gold on four worktrees simultaneously.
You don't want to have (0,1) skill developers and hand them these tools. You just can't afford it. Bad devs are now worse than ever while great ones spin gold on four worktrees simultaneously.
Reposted by atticus goldfinch
there's this interesting thing where certain pre-genai text sounds like genai and the common denominator is that it's a well-educated non-native speaker who is being excessively formal and/or kissing your ass
me: why does literally every AI tool glaze me, I'm just asking a question about documentation
friend: That's a really perceptive question - I think it gets at the heart of the current problem with LLMs! If you'd like, I can break down the origins and implications of this particular problem
friend: That's a really perceptive question - I think it gets at the heart of the current problem with LLMs! If you'd like, I can break down the origins and implications of this particular problem
December 8, 2025 at 4:08 AM
there's this interesting thing where certain pre-genai text sounds like genai and the common denominator is that it's a well-educated non-native speaker who is being excessively formal and/or kissing your ass
Reposted by atticus goldfinch
There is a real, fundamental misunderstanding at work here, and what I wish is that Matty would understand that many other people view an open racist and eugenicist with the same level of visceral contempt that he reserves for a left-wing writer who insulted his blog once.
I like these three. He’s also done a bunch of pro-vaccine stuff that I like to recommend to conservatives because but that wouldn’t be informative to the BlueSky audience.
December 8, 2025 at 2:30 AM
There is a real, fundamental misunderstanding at work here, and what I wish is that Matty would understand that many other people view an open racist and eugenicist with the same level of visceral contempt that he reserves for a left-wing writer who insulted his blog once.
Reposted by atticus goldfinch
Reposted by atticus goldfinch
jake tapper would set himself on fire outside of the white house to protest biden not retiring lmao
They’re clearly plugging this dude into an IV regularly and, if this were Biden, there would be demands for weekly hour long press conferences from his doctors on his health. But that’s not how authoritarianism works so stop asking questions already.
Trump last night had the complexion of a Thanksgiving turkey that was left in the oven too long and bandages were again visible on the back of his right hand
(photos via Getty and by Aaron Schwartz and Paul Morigi)
(photos via Getty and by Aaron Schwartz and Paul Morigi)
December 7, 2025 at 10:34 PM
jake tapper would set himself on fire outside of the white house to protest biden not retiring lmao
Reposted by atticus goldfinch
not sure how one argues with a straight face anymore that liberal elites did not play an outsized role in the slow-then-sudden disappearance of the cordon sanitaire around open racism and eugenics
Thanks for validating me, governor
December 7, 2025 at 10:14 PM
not sure how one argues with a straight face anymore that liberal elites did not play an outsized role in the slow-then-sudden disappearance of the cordon sanitaire around open racism and eugenics
Reposted by atticus goldfinch
I call this the Dream Vampire Economic Model. It also exists in the arts, publishing, and every other somewhat prestigious industry that young people want to work in.
there's a parallel between nonprofit world and academia, where there's a constant tide of idealistic, hopeful people coming in the door, and the people on top are able to lever that commitment to The Cause to extract from them in a way that's harder when everyone's just there to make a buck
December 7, 2025 at 6:18 PM
I call this the Dream Vampire Economic Model. It also exists in the arts, publishing, and every other somewhat prestigious industry that young people want to work in.
Reposted by atticus goldfinch
"Alabama should make the playoff they're clearly the 10th or 11th best team in the country" I'm sorry I remember when our country didn't loudly celebrate mediocrity. Participation trophy-ass national championship format - a sign of national decline.
December 7, 2025 at 5:14 PM
"Alabama should make the playoff they're clearly the 10th or 11th best team in the country" I'm sorry I remember when our country didn't loudly celebrate mediocrity. Participation trophy-ass national championship format - a sign of national decline.
Reposted by atticus goldfinch
Americans are angry because we're constantly being asked by elites to believe what we all know are lies, like "this team with two losses deserves a shot at the college football national championship." They don't. Facts don't care about your feelings.
December 7, 2025 at 5:13 PM
Americans are angry because we're constantly being asked by elites to believe what we all know are lies, like "this team with two losses deserves a shot at the college football national championship." They don't. Facts don't care about your feelings.