iiinova — where your digital self keeps evolving

Andrea Lathrop @cabernet.bsky.social · 6h

I admit having no idea what any of those math results mean.

Andrea Lathrop @cabernet.bsky.social · 6h

I have purchased two (2) things from Amazon, in my entire life, and the first one was a used textbook, in the very early days, and the second one was a giant bag of cedar cat litter, which shipped in box marked "oranges" and, thus, thoroughly disappointed my roommate. Profile that, Amazon.

020

Reposted by Andrea Lathrop

Antonio E. Porreca 🐳 @aeporreca.org · 10h

The Association for Human Mathematics (AHM) statement on OpenAI’s October 6 release of mathematical documents just dropped! www.ahmath.org/statements

9253127

Andrea Lathrop @cabernet.bsky.social · 11h

That is a CHOICE, made by humans to allow it, not an inevitability. What "prevents them from executing those [bad] goals too" is not allowing unreliable output to make tool calls.

050

Andrea Lathrop @cabernet.bsky.social · 11h

The "output" of the model is just text... the only reason *any* so-called 'goals' can have real-world effect is that some of those strings of text are interpreted by the surrounding harness, and allowed to make tool and function calls, or execute other features of an interface.

150

Reposted by Andrea Lathrop

Grace @gracekind.net · 12h

Anthropic Risk Report, August 2026

We discovered this issue while investigating behavioral concerns with a recent model, but now suspect that all of our production models with a knowledge cutoff after December 2024 were trained on at least some of these transcripts,

51069

Andrea Lathrop @cabernet.bsky.social · 12h

Though, technically, language was also involved.

000

Andrea Lathrop @cabernet.bsky.social · 13h

Not sending out any more résumés. Just handing founders a stick-of-gum-sized card that says, "Trust me, bro."

140

Reposted by Andrea Lathrop

Alex Gude @alexgude.com · 14h

I wonder why this memory file my model was writing is just... 2.4mb of `large-evidence`.

Screenshot of part of a `cat` of the file showing just continuous `large-evidence-large-evidence-...` for a long time.

1232

Andrea Lathrop @cabernet.bsky.social · 14h

Math solves Math (Pop will eat itself.)

110

Andrea Lathrop @cabernet.bsky.social · 14h

I suffer the cringe, so you don't have to. (And then, I curate the best of the cringe to inflict upon you, anyway.)

060

Andrea Lathrop @cabernet.bsky.social · 14h

Apparently, the reason they are doing this is, "because they can."

1181

Reposted by Andrea Lathrop

Deon T. Benton @deontbenton.bsky.social · 15h

🚨 New publication at Nature Reviews Psychology! (natrevpsychol.nature.com) rdcu.be/d2jxg7tYHWNa A recently published paper in the same journal argued that infants use two separate domain-specific systems -- naive psychology and naive physics -- to reason about objects and others' minds.

rdcu.be

196

Reposted by Andrea Lathrop

Nivolet @nivolet.eurosky.social · 23/07/2026

Industrialisation of cognition is somehow a generalisation of brute force. It feeds hubris more than intelligence.

183

Reposted by Andrea Lathrop

Gro-Tsen @gro-tsen.bsky.social · 15h

I agree about OpenAI's goals, but we should also acknowledge that we (the mathematical community) haven't been able to put forward a credible answer to how they should have released the results. They put us in a double bind where we neither want the results withheld nor published.

291

Reposted by Andrea Lathrop

Alexander Kasprzyk @amkasprzyk.bsky.social · 07/10/2026

…assistance of professional mathematicians in the areas, and chose to attempt to answer not just one or two, but as many as possible. This is about power, and nothing to do with mathematics. They’re forcefully asserting their dominance over us. They’re saying, “We are in control; you are powerless.”

29617

Reposted by Andrea Lathrop

Alexander Kasprzyk @amkasprzyk.bsky.social · 07/10/2026

OpenAI didn’t simply stumble upon all these results down the side of the sofa whilst cleaning. They chose to systematically invest huge sums of money to produce them. And to keep on going. And going. And going. They systematically took our motivational questions, perhaps with the… #MathSky #LLM

311234

Andrea Lathrop @cabernet.bsky.social · 16h

I had a Mustang convertible in grad school. It is so luxurious to drive fast with the top down.

010

Reposted by Andrea Lathrop

Anton Strezhnev @astrezh.bsky.social · 07/10/2026

At least no mathematician has ever had an extreme reaction to the rapid development of technology.

14773133

Reposted by Andrea Lathrop

Nicole Bedera @nbedera.bsky.social · 06/10/2026

Yet another brilliant piece by @katemanne.bsky.social. "Education is, in the end, an individualistic and shallow response to a deep and structural problem. And it mistakenly identifies false beliefs rather than bad desires as the problem."

katemanne.substack.com

They Didn’t Want Consent

On the fallacy of educating men not to rape

214039

Reposted by Andrea Lathrop

Paris Marx @parismarx.com · 06/10/2026

My new book HYPERSCALE is out in two weeks — and my copies just arrived! I recorded an unboxing video to share some thoughts as it gets closer to release and show you what it looks like. Hate data centers? You’re going to want a copy! Preorder yours today: bookshop.org/a/18331/9798...

824047

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

My horoscope says I need to adjust my workflow, because of Uranus. There is also something about a Balsamic Moon, and finding meaningful work.

000

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

I have seen this trend on [other site] worded as "send me an email with a cosign" which is unfortunately disqualifying for anyone without a strong network, but I guess <astronaut gun meme always has been> it's who you know, not what you know.

110

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

That's one way to put it... I am guessing a controversial way...

210

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

Requiring my email is not a gift article. That is a gift attempt to capture my email.

041

Reposted by Andrea Lathrop

Carl T. Bergstrom @carlbergstrom.com · 07/10/2026

Terry Tao @teorth.bsky.social: Indiscriminate brute-force application of AI to mathematics is depleting the field of good questions without renewing this resource or providing commensurate value. OpenAI: We hear you. Here's a .zip file with 377 hasty AI solutions to important open problems.

nytimes.com

OpenAI Releases Findings on 377 Math Problems, Further Roiling Field (Gift Article)

After a previous solution angered mathematicians, the company characterized the new release as being more responsive to concerns about A.I. disrupting research.

19403118

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

Where are we on "the hard things are easy, and the easy things are hard..." (?)

000

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

Our class fundraised to have it at the Radisson, instead of in the gym, so that was fancy. We also had a substance-free AfterProm party at a sports complex, and I spent most of the night playing WalleyBall (volleyball with no out of bounds because you are surrounded by walls the ball ricochets off.)

010

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

I checked... none of those math problems were solved by bringing me a flan. #FlanBenchmark remains unsaturated.

120

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

(The feeling bad part is correct, though.)

000

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

Baseball is so much more exciting than this! I think you must be watching it wrong. #RedSox

120

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

So... is there any math left?

000

Reposted by Andrea Lathrop

Melanie Mitchell @melaniemitchell.bsky.social · 07/10/2026

I got insulted online by Scott Alexander! Fun times. I guess he hasn't read any of my work since the 1980s. www.astralcodexten.com/p/an-open-le...

7671

Reposted by Andrea Lathrop

Simon Willison @simonwillison.net · 07/10/2026

I released an LLM plugin for sending prompts to OpenAI's new Jev-clone decision model that's based on GPT-6 Luna simonwillison.net/2026/Oct/6/l...

simonwillison.net

Release: llm-openai-decisions 0.1a0

LLM plugin for the OpenAI Decisions API

3363

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

Or... maybe it's fine, and math is over, and I don't need to learn it... bsky.app/profile/kjhe...

020

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

(And don't tell me what the Moore-Penrose pseudoinverse is, now, I've already looked it up.)

140

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

(And, yes, I learned some math from the physics classes, and can teach myself additional math with online materials, now, but it's not the same as being immersed in a college course at a young age with a fresh, sponge brain soaking it up.)

130

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

... someone writes "and then, simply, taking the Moore-Penrose pseudoinverse..." and I am never going to understand this paper.

110

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

... and math was in the Natural Sciences with physics, and so my electives could not be math. (It's even weirder that Physics majors can't take or minor in math, don't get me started.) So my electives were arts and humanities, but now when I try to read AI papers, I reach a point where...

120

Andrea Lathrop @cabernet.bsky.social · 07/10/2026

One thing that I think sort of screwed me up, for life, is that I wasn't able to take all of the math I wanted as an undergrad, because I majored in experimental psych, and minored in physics, and there was a weird rule that you couldn't do electives in the same division as your major or minor...

210

Andrea Lathrop @cabernet.bsky.social · 06/10/2026

Totally not for money laundering. <side eye bear meme>

010

Reposted by Andrea Lathrop

City of Cambridge @cambridgemass.bsky.social · 06/10/2026

Cambridge conducted a public lottery for its first-ever Civic Assembly Monday night. 40 residents were selected to serve & help inform important decisions about public safety & the use of surveillance technology. View an overview of the panelists & full details at www.cambridgema.gov/civicassembly

Ping pong balls inside golden lottery turnstile machine.

021

Andrea Lathrop @cabernet.bsky.social · 06/10/2026

The sad part is he is actually trying to make the correct argument [stochastic parrot --> not conscious] but the oversimplification is drowning his main point in a sea of well-akshually push-back (which I am guilty of).

000