Jeff Lockhart

Went to another faculty meeting with a vote because I've got to get democracy somewhere.

"AI can make up fake data so I get one more research paper", oh how small minded.

Photograph of an orange cat beside a laptop

Adorno had some words on this...

Lol at the man on BART asking people to sign a petition against raising property taxes because "it's good if you want to own a house some day."

You'd also bring it back a week later

Okay yes but the analytic philosophers who happily define it aren't good either.

I need a browser extension that keeps all the people cat posting, but translates their cutesy "cat speak" typing into standard English.

Harder to notice may be more subtle things, like possibly inserting "not" to reverse meaning where someone said something deliberately unusual (common to Lewis Carrol, but also plenty of academic theorists).

Newer generative models have a different design that can introduce bullshit/hallucinations/whatever you want to call things not in the original text. Some examples already out there with, e.g., inserting "translated by [name]" from training data.

I think there's a shift in the underlying tech for automated translation that poses an issue. Previous work was designed for translation specifically, with a sort of 1-1 map between sentences approach. (Same for speech to text.)

Is there money or even a submission track for those of us who want to do anything with AI other than fabricate data? no. Of course not.

Lol there's an academic conference on how to fabricate data with AI, and it's all-expense-paid for everyone accepted to present, because of course it is.

Ack! So many thoughts. This is useful and generative.

This helped me distinguish between the delusional anthropomorphism that bugged me and a similar not-bad thing. Since it came out, a few more papers have been written arguing back and forth over chatbot fictionalism.

journals.publishing.umich.edu/ergo/article...

Have you read the stuff on "fictionalism"?

Adobe when I want to read a PDF: 📎 here are 10 pop ups about our new AI features

Adobe when I need to figure out how to use a niche technical feature of calculated fields in PDF forms: 🦗🦗

Hoping for a story like the iron in spinach decimal place slip.

A digital CAPTCHA verification window titled "Select all squares with PIPES" against a plain white background. The window contains a 3×3 grid of numbered squares, mixing literal hardware, smoking pipes, and programming syntax.

These captchas just keep getting harder #rstats

screenshot of the text: "Key Highlights of the partnership Refreshed Cal Student Store Experience We are investing in a modernized store environment with updated layouts, improved flow, and curated assortments that reflect UC Berkeley’s culture and values."

I understand associate deans want to feel important, but I have no desire to get emails about this.

What a pairing

We're glad to have you!

Watching another person who studies the far right slip into the both-sides centrism of 'polarization.' Someone with tenure should write about this pipeline.

Kevin, please, it's my favorite thing: pathological scholarship.

wait lol this would actually explain some things

Grim.

"you're listening to people fixing the world from BBC world service..." This is a wild experience.

Don't want to quote dunk a rando, but I saw someone criticizing a new data center plan because "AI wastes water" ... except the plan they linked to doesn't use any evaporative cooling. It even says "dry coolers" on the cover image of the link. This stylized fact is way out of hand.

The persona selection model A theory of why AI models act like humans

This week in "anthropic discovers what the porn subreddits have known for 4 years," www.anthropic.com/research/per...