Hodgetwins Mike Solana Twitter acting by itself to suppress free speech is not a 1st amendment violation, but acting under orders from the government to suppress free speech, with no judicial review, is

We need a new national holiday: Dec. 23- Email Bankruptcy Day. On this day, all unread, unanswered emails are annulled. It is an email jubilee, as if all those messages were never sent. No one sends an email until Jan. 2, and we all start the new year inbox zero.

53. The thing that makes moral disagreement with friends feel so unsafe is the thing I was talking about earlier: People are on a hair trigger for any hint of it, which means you can't even broach the possibility of the subject.

Lots of Research, No Evidence pershmail.substack.com/p/lots-of-rese…

"GPT-2 memorizes content that has since been removed from the Internet, and is thus now primarily accessible through GPT-2... LMs that are trained entirely on (at-the-time) public data may end up serving as an unintentional archive for removed data." arxiv.org/abs/2012.07805

Output Parsers (extracting a structured response from LLM output) sometimes fail. So what do you do then? You ask the LLM to fix it's output of course! Introducing Output Parsers that can fix themselves (OutputFixingParser, RetryOutputParser) 👇

Here are the three most radical policy changes I'm suggesting. If they were implemented and I turned out to be wrong about social media and smartphones, how bad would it be? And if I'm right and we don't do them?

“We show that LaTeX users were slower than Word users, wrote less text in the same amount of time, and produced more typesetting, orthographical, grammatical, and formatting errors. “ journals.plos.org/plosone/articl… Thanks Florian Buehler

Coming from an MIT undergrad experience, this original paper raised several suspicions. This response confirms many of them. "As the field progresses faster and faster, timelines for discovery seem to shrink, which often comes with shortcuts..."

If you’re looking for a roommate for #ICER2023, here’s a spreadsheet to help with roommate matching thanks to Elijah Rivera: docs.google.com/spreadsheets/d…

We used to run the airline industry on the (reasonable, at the time)assumption that nobody wanted to blow up the plane. Sadly that didn't work, so now we make people prove that they aren't going to blow up the plane. I think we need to run scientific publishing like this. /1

Academics in computer science: great talk! I would love to do interdisciplinary work myself. Also academics in computer science: I don't see her technical contribution. Next candidate?

“Where are the PHYSICAL SHUTTERS on my electronics’ built in cameras? Where are the PHYSICAL KILl SWITCHES to the microphones? Why is the camera indicator light controlled by software that is WEB ACCESSIBLE?” Been asking these questions from *DAY 1* of the smartphone/laptop era.

I'm looking for a tool that can take mini-scripts for children (think early childhood "play kitchen") and produce corresponding images, almost WikiHow style. Which generative AI tools already do this? And/or how easy is this to create myself?

Formal methods people: we're introducing a new track for papers at FM: TUTORIAL PAPERS. Here is the CFP. Please share with anyone who you think would have useful things to say! I want to grow the quality of FM education. Do ask if you have questions!

Crazy AF. Paper studies AK and Aran Komatsuzaki paper tweets and finds those papers get 2-3x higher citation counts than control. They are now influencers 😄 Whether you like it or not, the TikTokification of academia is here! arxiv.org/abs/2401.13782

Something to read for the weekend: link.springer.com/epdf/10.1007/s…

Had some fantastic days at #ITiCSE 2024 in Milan and met great researchers/educators with inspiring ideas. I'm also glad that I could present our work [Elijah Rivera, Kathi Fisler and Shriram Krishnamurthi (primary: Bluesky)] about iterative problem planning using transformers.

We're doing qualitative coding. For each top-level code in a category, we also select multiple modifier tags (think "pos/neg sentiment", "examples yes/no", etc.). The end result is exponential blowup of code combinations. What is an appropriate measure of inter-coder reliability?

I'm super excited to experiment with enabling programmers to define their own event hat blocks in Snap! Cloud, expressed as predicates describing the condition which fires the event: