Andrea Lathrop
@cabernet.bsky.social
Cognitive Science... tech stuff... dev psych... AI... 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
(Am I missing the links to the details about Jev/System One, or are there not any details? The blog looks sparse and vague.)
I don't know if it's the 'personalist form of government' driving this, or merely that Donald has a storied history (Bonwit Teller frescoes, the East Wing, the Reflecting Pool...) of destroying publicly valuable things for private gain.
One aspect of living in a personalist form of government is that people no longer expect the laws or norms to protect them. If the king is rumored to be thinking of demolishing a believed institution, people assume they must defend it in person
September 19, 2026 at 7:40 PM
I don't know if it's the 'personalist form of government' driving this, or merely that Donald has a storied history (Bonwit Teller frescoes, the East Wing, the Reflecting Pool...) of destroying publicly valuable things for private gain.
We live in the dumbest timeline.
September 19, 2026 at 7:34 PM
We live in the dumbest timeline.
I made two stupid mistakes, and stopped in the middle to wash two loads of dishes, but I got it... eventually...
September 19, 2026 at 7:13 PM
I made two stupid mistakes, and stopped in the middle to wash two loads of dishes, but I got it... eventually...
Idly wondering if the press ban is directly related to the upcoming meeting with Xi... specifically, if the optics of it is something Trump thinks will impress Xi.
September 19, 2026 at 5:04 PM
Idly wondering if the press ban is directly related to the upcoming meeting with Xi... specifically, if the optics of it is something Trump thinks will impress Xi.
Reposted by Andrea Lathrop
has anyone seen a good write up of how the “frontier labs” (vom) are training their models to hack before they oops! leave them unsecured and unmonitored and they - surprise! - hack?
September 19, 2026 at 2:27 PM
has anyone seen a good write up of how the “frontier labs” (vom) are training their models to hack before they oops! leave them unsecured and unmonitored and they - surprise! - hack?
My Twitter feed is David Sacks telling AI companies no law prevents them from colluding on a safety slow-down, and my Bsky feed is a California class-action suit claiming coordinating a slowdown is illegal collusion.
September 19, 2026 at 2:43 PM
My Twitter feed is David Sacks telling AI companies no law prevents them from colluding on a safety slow-down, and my Bsky feed is a California class-action suit claiming coordinating a slowdown is illegal collusion.
You really have to wonder who decided that the tech at the time was reliable enough to use it this way, and who, and how many people in the chain were warned about how unreliable it actually is, and deployed it, anyway.
US officials say overreliance on Palantir's Maven AI system was among the factors that contributed to a February missile strike in Iran that killed 123 children (Bloomberg)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
September 19, 2026 at 2:16 PM
You really have to wonder who decided that the tech at the time was reliable enough to use it this way, and who, and how many people in the chain were warned about how unreliable it actually is, and deployed it, anyway.
Reposted by Andrea Lathrop
The Kennedy Center was financially stable until Trump took over: "In the fiscal year ending in 2024, when the center was led by President Deborah Rutter and a bipartisan board of trustees, it brought in about $40 million more than it spent"
www.theatlantic.com/culture/2026...
www.theatlantic.com/culture/2026...
What the Kennedy Center’s Tax Return Reveals
A previously unseen filing captures the state of the institution seven months after Donald Trump’s takeover.
www.theatlantic.com
September 19, 2026 at 10:53 AM
The Kennedy Center was financially stable until Trump took over: "In the fiscal year ending in 2024, when the center was led by President Deborah Rutter and a bipartisan board of trustees, it brought in about $40 million more than it spent"
www.theatlantic.com/culture/2026...
www.theatlantic.com/culture/2026...
Reposted by Andrea Lathrop
As someone who was working on a cognitive psychology Ph.D. in the 90s, long before these shits took over, it pisses me off that this term was ruined by them
Related, annoying white guys renamed bias "priors" as if they were doing math
it's funny but also a reminder that the fundamental premise here is that a Black man can't be objective when talking about race while a white man can because Black people do identity politics and white people do facts and this my friends is what I have named "white empiricism"
September 19, 2026 at 1:13 PM
As someone who was working on a cognitive psychology Ph.D. in the 90s, long before these shits took over, it pisses me off that this term was ruined by them
Reposted by Andrea Lathrop
My Twitter timeline says it's all a psy-op, and my Bsky timeline says it's not a psy-op.
September 19, 2026 at 2:00 PM
My Twitter timeline says it's all a psy-op, and my Bsky timeline says it's not a psy-op.
I cannot repeat this enough. I'm seeing "haha, starched shirt consultants" takes on the Anthropic announcement of the Accenture embed, BUT Accenture acquired 'Faculty' in January, which originated as 'ASI Data Science' founded with financial backing from Jaan Tallinn. Just as like-minded as METR.
September 19, 2026 at 12:39 AM
I cannot repeat this enough. I'm seeing "haha, starched shirt consultants" takes on the Anthropic announcement of the Accenture embed, BUT Accenture acquired 'Faculty' in January, which originated as 'ASI Data Science' founded with financial backing from Jaan Tallinn. Just as like-minded as METR.
Reposted by Andrea Lathrop
My understanding of AI, as a software engineer, is that a complicated statistical model won't suddenly get sentient and decide to kill us; but people who consider AI a magic authority will use it irresponsibly in high-stakes decisions, and those people *may* kill us
so we just nearly started a war with china (a nuclear state) because of genAI use
September 18, 2026 at 9:57 PM
My understanding of AI, as a software engineer, is that a complicated statistical model won't suddenly get sentient and decide to kill us; but people who consider AI a magic authority will use it irresponsibly in high-stakes decisions, and those people *may* kill us
Interesting... so... I just did a quick online search, and in Jan 2026, Accenture acquired 'Faculty' which was founded in 2012 as 'ASI Data Science' (backed by Jaan Tallinn) and rebranded in 2019, after having messy ties to Cambridge Analytica...
Anthropic named Accenture's Faculty division as its first embedded AI evaluator, with the companies planning to invest $1 billion over five years
Anthropic's first embedded evaluator is ... Accenture? | TechCrunch
Accenture is about to take on its most high-risk consulting engagement ever.
techcrunch.com
September 18, 2026 at 11:00 PM
Interesting... so... I just did a quick online search, and in Jan 2026, Accenture acquired 'Faculty' which was founded in 2012 as 'ASI Data Science' (backed by Jaan Tallinn) and rebranded in 2019, after having messy ties to Cambridge Analytica...
People (including me) are posting too much about x-risk, and too little about xenocortical (XCX) mice...
September 18, 2026 at 9:40 PM
People (including me) are posting too much about x-risk, and too little about xenocortical (XCX) mice...
Building the bio lab from the t-shirt "Don't give AI autonomous control of a bio lab!"
September 18, 2026 at 7:08 PM
Building the bio lab from the t-shirt "Don't give AI autonomous control of a bio lab!"
Reposted by Andrea Lathrop
Drop everything @felixsalmon.com wrote about the $60 bread www.bloomberg.com/news/article...
The K-Shaped Loaf of Bread
At RYE by Martin Auer, the bread costs $60 per loaf. The theory behind this gambit makes some sense.
www.bloomberg.com
September 18, 2026 at 4:42 PM
Drop everything @felixsalmon.com wrote about the $60 bread www.bloomberg.com/news/article...
Reposted by Andrea Lathrop
To me there's "AI safety, the fantasy realm" and then there are actual serious people working in actual AI safety and (yes) "alignment" even if that term is tainted. The problem is that they often all get lumped together. And too many top people are obsessed with the fantasy realm.
September 17, 2026 at 4:44 AM
To me there's "AI safety, the fantasy realm" and then there are actual serious people working in actual AI safety and (yes) "alignment" even if that term is tainted. The problem is that they often all get lumped together. And too many top people are obsessed with the fantasy realm.
Good morning, I have had two (2) hours of sleep. And one of those was *after* I took the trash to the curb, at 8:15 a.m.
September 18, 2026 at 1:27 PM
Good morning, I have had two (2) hours of sleep. And one of those was *after* I took the trash to the curb, at 8:15 a.m.
Reposted by Andrea Lathrop
Well, that didn’t take long. And its probably like this - Jev is an LLM based on diffusion, trained to the max on synthetic classification tasks. Without reasoning and tools, what we have is a fast cheap and often dumb predictor. For which I’d still like to see real use cases.
Most convincing one so far from @mmastrac.bsky.social ! x.com/mmastrac/sta...
September 18, 2026 at 5:57 AM
Well, that didn’t take long. And its probably like this - Jev is an LLM based on diffusion, trained to the max on synthetic classification tasks. Without reasoning and tools, what we have is a fast cheap and often dumb predictor. For which I’d still like to see real use cases.
Reposted by Andrea Lathrop
a hill i will die on is that people realy really don't understand just how much stuff is in the training data
September 18, 2026 at 4:27 AM
a hill i will die on is that people realy really don't understand just how much stuff is in the training data
Reposted by Andrea Lathrop
Great piece by @xriskology.bsky.social on how the effective altruism cult has helped destroy actual philanthropy - with dangerous and deadly results for many people in need:
Effective Altruism Is a Dangerous Cult — Here's Why
Understanding the "Scientology of Silicon Valley." (7,700 words)
www.realtimetechpocalypse.com
September 17, 2026 at 1:43 PM
Great piece by @xriskology.bsky.social on how the effective altruism cult has helped destroy actual philanthropy - with dangerous and deadly results for many people in need: