Andrea Lathrop
@cabernet.bsky.social
Wine... Cognitive Science... Photography... tech stuff... former Aslin Lab (Master's) and Eppler Lab (with Jackie Gibson emeritus) so I'm a weird mix of Gibsonian affordances, visual statistical learning and anticipatory eye movements. Sometimes academic.
Pinned
I'm so mad.
Just to make this abundantly clear:
February 16, 2026 at 6:30 PM
Just to make this abundantly clear:
Idly noticing that Jeffrey Epstein was obsessed with overpopulation, while Elon is obsessed with underpopulation.
February 16, 2026 at 4:47 AM
Idly noticing that Jeffrey Epstein was obsessed with overpopulation, while Elon is obsessed with underpopulation.
"... Whenever you want to speak with someone, you could first interact with their agentic harness..."
February 16, 2026 at 2:07 AM
"... Whenever you want to speak with someone, you could first interact with their agentic harness..."
Could we at least have the flapper dresses and the Charleston?
This is where I'm at on the last few years, and I think it's become clear just how much the 1920s were a product of what influenza wrought
The world went insane in the decade immediately after the last global respiratory pandemic of their type (Spanish flu), I think there's something about this kind of collective trauma in modern societies that had managed to reduce that kind of mass mortality that does this to politics.
February 16, 2026 at 1:05 AM
Could we at least have the flapper dresses and the Charleston?
Reposted by Andrea Lathrop
sorry that you threw an uncool enough party though that your friend had to pretend to go home to feed his LLM pet rock
February 16, 2026 at 12:54 AM
sorry that you threw an uncool enough party though that your friend had to pretend to go home to feed his LLM pet rock
"... strategic AI talent move..."
📰 OpenClaw Creator Peter Steinberger Joins OpenAI in Strategic AI Talent Move
OpenAI has recruited Peter Steinberger, the creator of the open-source AI framework OpenClaw, in a significant move to bolster its advanced reasoning and agent-based AI capabilities. The acquis...
#AINews #AI #Teknoloji
OpenAI has recruited Peter Steinberger, the creator of the open-source AI framework OpenClaw, in a significant move to bolster its advanced reasoning and agent-based AI capabilities. The acquis...
#AINews #AI #Teknoloji
OpenClaw Creator Peter Steinberger Joins OpenAI in Strategic AI Talent Move
OpenAI has recruited Peter Steinberger, the creator of the open-source AI framework OpenClaw, in a significant move to bolster its advanced reasoning and agent-based AI capabilities. The acquisition signals a shift toward open collaboration in next-generation AI development.
aihaberleri.org
February 16, 2026 at 12:36 AM
"... strategic AI talent move..."
(Oh, good, more venture signaling coworker tweets.)
February 15, 2026 at 11:48 PM
(Oh, good, more venture signaling coworker tweets.)
Reposted by Andrea Lathrop
Lol
OpenClaw founder Steinberger joins OpenAI, open-source bot becomes foundation reut.rs/4auGm8A
OpenClaw founder Steinberger joins OpenAI, open-source bot becomes foundation
Peter Steinberger, the founder of OpenClaw, is joining OpenAI, and the open-source bot is becoming a foundation, OpenAI CEO Sam Altman said on Sunday.
reut.rs
February 15, 2026 at 10:41 PM
Lol
<Resisting the urge to jump into more esoteric conversations about machine consciousness.>
February 15, 2026 at 10:58 PM
<Resisting the urge to jump into more esoteric conversations about machine consciousness.>
Reposted by Andrea Lathrop
douglas adams was our most accurate futurist
We are at an airport restaurant. A robot just drove up to us with our food, said, "Hi! Here I am!" and then drove away with our food.
February 15, 2026 at 10:10 PM
douglas adams was our most accurate futurist
? 🧐
Microbiome depletion rejuvenates the aging brain https://www.biorxiv.org/content/10.64898/2026.02.13.705770v1
February 15, 2026 at 8:29 PM
? 🧐
Probably if I stopped typing all of my opinions into this computer, I would have fewer notification bells to respond to, every time I return and open it.
February 15, 2026 at 8:06 PM
Probably if I stopped typing all of my opinions into this computer, I would have fewer notification bells to respond to, every time I return and open it.
If this is the dichotomy, I am definitely a physicalist.
Physicalists aspire to give an account of how consciousness emerges from physical reality.
Panpsychists aspire to give an account of how physical reality emerges from consciousness.
One of these explanatory projects has been more successful than the other. Can you guess which one?
Panpsychists aspire to give an account of how physical reality emerges from consciousness.
One of these explanatory projects has been more successful than the other. Can you guess which one?
How does the "explain everything else" bit work?
February 15, 2026 at 4:03 PM
If this is the dichotomy, I am definitely a physicalist.
Your face data is only as secure as the amount of Adderall your friends are on when they update their contact photos...
Just spent two hours updating the contact photos for everyone I've texted in the last month and I don't know if that means the adderall is working or not working.
February 15, 2026 at 3:56 PM
Your face data is only as secure as the amount of Adderall your friends are on when they update their contact photos...
I have to stop falling for 'gotcha's that are largely beside the point.
February 15, 2026 at 3:50 PM
I have to stop falling for 'gotcha's that are largely beside the point.
Reposted by Andrea Lathrop
LLMs have learned to generalize, which is why I hired an army of subject matter experts to tediously write out custom reasoning texts on every possible problem so that they never have to generalize
February 15, 2026 at 3:34 PM
LLMs have learned to generalize, which is why I hired an army of subject matter experts to tediously write out custom reasoning texts on every possible problem so that they never have to generalize
Reposted by Andrea Lathrop
"I've hooked the synthetic text extruding machine up to other systems that take that text as input and impact the world" != "The synthetic text extruding machine knows how to use tools."
February 15, 2026 at 4:43 AM
"I've hooked the synthetic text extruding machine up to other systems that take that text as input and impact the world" != "The synthetic text extruding machine knows how to use tools."
Reposted by Andrea Lathrop
Short musings on "cognitive debt" - I'm seeing this in my own work, where excessive unreviewed AI-generated code leads me to lose a firm mental model of what I've built, which then makes it harder to confidently make future decisions simonwillison.net/2026/Feb/15/...
How Generative and Agentic AI Shift Concern from Technical Debt to Cognitive Debt
This piece by Margaret-Anne Storey is the best explanation of the term cognitive debt I've seen so far. Cognitive debt, a term gaining traction recently, instead communicates the notion that …
simonwillison.net
February 15, 2026 at 5:22 AM
Short musings on "cognitive debt" - I'm seeing this in my own work, where excessive unreviewed AI-generated code leads me to lose a firm mental model of what I've built, which then makes it harder to confidently make future decisions simonwillison.net/2026/Feb/15/...
Reposted by Andrea Lathrop
Sort of gives away the lie that they're arresting hardened criminals who are drags on society when they're setting up traps to lure in Good Samaritans
CCTV footage appears to show DHS agents posing as people having car trouble in order to lure a target out of his house before arresting them. The target comes out and starts helping them before 3 vehicles surround the scene. The car "experiencing" trouble then drives off with them.
February 15, 2026 at 1:10 AM
Sort of gives away the lie that they're arresting hardened criminals who are drags on society when they're setting up traps to lure in Good Samaritans
Given that 1/3 of respondents have never heard of Pete Hegseth, I'm wondering what portion of RFK Jr's 'favorable' ratings are from people who thought they were rating his father?
Americans see several key Trump administration officials more negatively than positively.
(Pew Research Center)
More, via Opinion Today:
opiniontoday.substack.com/p/260213-top...
(Pew Research Center)
More, via Opinion Today:
opiniontoday.substack.com/p/260213-top...
February 14, 2026 at 8:51 PM
Given that 1/3 of respondents have never heard of Pete Hegseth, I'm wondering what portion of RFK Jr's 'favorable' ratings are from people who thought they were rating his father?
Persona was the company Rev used, too, when they forced it upon the transcriptionists. The roll-out was submit, or quit.
turns out the discord situation is worse! the company that will be doing the age verifications with IDs is Persona, a peter thiel aligned company
we're going to try to wind down the Kill The Computer discord and try to get a subreddit for the podcasts community up and running
we're going to try to wind down the Kill The Computer discord and try to get a subreddit for the podcasts community up and running
February 14, 2026 at 8:20 PM
Persona was the company Rev used, too, when they forced it upon the transcriptionists. The roll-out was submit, or quit.
Here is what I keep coming back to, when I read things like this: What is this 'non-human economy' that will be 'fully sustained' by AI? Economies are human tools. What is an economy to an algorithm? Once everything is fully automated, why is anything still measured in terms of an economy?
February 14, 2026 at 7:44 PM
Here is what I keep coming back to, when I read things like this: What is this 'non-human economy' that will be 'fully sustained' by AI? Economies are human tools. What is an economy to an algorithm? Once everything is fully automated, why is anything still measured in terms of an economy?
If you are claiming "genuinely new knowledge" output from a LLM, the burden is on you to prove that said knowledge exists absolutely nowhere in the training data, and could not have leaked in during any other part of your methodology.
February 14, 2026 at 6:10 PM
If you are claiming "genuinely new knowledge" output from a LLM, the burden is on you to prove that said knowledge exists absolutely nowhere in the training data, and could not have leaked in during any other part of your methodology.