Ever thought we acquire generalizable knowledge by discarding details and compressing our experiences?
In a new BBS paper, @sabinasloman.bsky.social and I argue otherwise, proposing a novel way of studying human learning inspired by double descent in ML.
Disagree? Propose a commentary by May 15 :)
Welcome to Manuscript. An imagined product. Inspired by Claude Design, mocked up for academic research. If Manuscript existed, I could open it and find every part of a study in one place. The instruments I built — surveys, protocols, consent forms. The papers I read in the run-up. The data those instruments collected. The figures I drew from it. And at the center of it all, the paper itself. What follows is a walk through what that might look like. None of it works.
The release of #ClaudeDesign left me wondering what a tool like this could look like for academics (rather than designers)
I then used Claude Design to mock it up for me
Let me show you what came of it... 🧪🤖🧵
rafaelmbatista.com/manuscript/i...
'This is Fine' meme. A cartoon dog wearing a small brown hat sits calmly on a chair at a table with a coffee mug. The room around the dog is completely engulfed in bright yellow and orange flames, with thick dark smoke rolling across the ceiling. White text at the top reads 'Everything is Fine'. Hanging on the green wall behind the dog is a framed picture of a green tree with a brown trunk against a light blue sky.
Something really nice about psychology studies conducted in a lab where everything is controlled
Tightly controlled psych studies be like ‘Adding picture of nature to your desk, increases life satisfaction, productivity, and job prospects (p’s<.05)’
I mean… it certainly won’t hurt
First person I thought of (amongst presidential hopefuls) is @petebuttigieg.bsky.social
Because I was all in for Pete but listening to some of his interviews this past year, he was never clear on what *exactly* he wanted to see done.
Dem Leaders need vision. I agree w/ Mamdani
Neat article that traces sources of “reasoning” discovery to… #4chan!
Whatever advantage OpenAI may have had, it would’ve been limited
www.theatlantic.com/technology/2...
🚨 U Melbourne flagship McKenzie Postdoc
- 3 years fully funded, no teaching
- 25K research budget
- fame and prestige
EOI closing 7 May, need to have a supervisor (email/DM me with Qs!)
sites.research.unimelb.edu.au/research-fun...
Lots of takes on this popping up on my feed. Many of them clearly unimpressed. Which feels kind of sad, like if things like this don't impress you, what does?
Folks must be watching #ArtemisII videos and thinking to themselves, "so what, I can see the moon from my window, not that impressive"
TFW you look out the window and see the Earth disappear
🎥 Astronaut Reid Wiseman
📍 Artemis II from the far side of the Moon
What a nice story about this learning process and this beautiful yet unconventional human-octupus interaction!
youtu.be/PcWnQ7fYzwI?...
Omg 😱 @corybooker.com is going to run a different campaign this time around. Excited to see it! 🔥
These incredible Buff Tip Moths blend in so well with branches that you might mistake them for sticks. Nature's masters of disguise are truly fascinating!
None of this exists. But it could! The tools for each component have already been deployed, they just don't live in the same ecosystem.
I've sketched out the idea, now someone just needs to get to work building it 😉
rafaelmbatista.com/manuscript/i...
#AI #AcademicSky
Finally, the Paper itself, displayed in three views. The Draft is the workbench where you write and revise. Publication previews a polished version for the journal you choose. Interactive makes the paper come to life.
Try it here: rafaelmbatista.com/manuscript/s...
Figures would be easily rendered. With buttons that allowed you to visualize your data differently. The code could be rewritten in seconds thanks to Claude Code. Toggle from a bar graph to a box plot and whats in the manuscript changes too.
Play with it here: rafaelmbatista.com/manuscript/s...
Data includes both raw versions and cleaned. Raw data is encrypted, personal details flagged and protected. Cleaned data is easily shared and cleaning scripts would be stored for reproducibility.
🔗 rafaelmbatista.com/manuscript/s...
The Library would store papers and excerpts for you to use. It would have graphic features we see today in Obsidian and NotebookLM, but it'd go further too, nudging researchers to cite specific sections of specific papers.
🔗 rafaelmbatista.com/manuscript/s...
The Materials section would store survey instruments, interview protocols, consent forms, and pre-analysis plans.
Everything would be timestamped and versioned.
🔗 rafaelmbatista.com/manuscript/s...
I started with a prompt where I imagined some essential features: a library for papers + other sources, a place to keep materials related to my studies, data files (separated raw vs clean) + the ability to turn that data into figures, etc
Here are the wireframes:
rafaelmbatista.com/manuscript/w...
Welcome to Manuscript. An imagined product. Inspired by Claude Design, mocked up for academic research. If Manuscript existed, I could open it and find every part of a study in one place. The instruments I built — surveys, protocols, consent forms. The papers I read in the run-up. The data those instruments collected. The figures I drew from it. And at the center of it all, the paper itself. What follows is a walk through what that might look like. None of it works.
The release of #ClaudeDesign left me wondering what a tool like this could look like for academics (rather than designers)
I then used Claude Design to mock it up for me
Let me show you what came of it... 🧪🤖🧵
rafaelmbatista.com/manuscript/i...
This release blew me away 🤯 Sunk a few hours this weekend playing around with it.
Wrote some of it up here:
rafaelmbatista.com/notes.html#3
There was a time when the president inspired pride instead of exhaustion. I miss that more than I can say.
I know nothing of physics, but in Econ they’ve also grappled w/ field becoming too math-y. Several have argued it went too far (unnecessarily) & something was lost.
If it feels like a req, it also risks alienating people who dislike the math but are good at other parts of the research.
Yes, and… those working on these “too technical” papers should also be taught to communicate w/o equations to a broader audience. Clear communication does matter. I don’t see it as dumbing down. In fact, it’s the communication that enables rigorous work w/o needing to simplify the methods
The FBI guy is on a paranoid bender and won’t leave his room.
The War guy quotes bible passages from Pulp Fiction.
The Health guy collects raccoon dicks.
It’s an Idiocracy on steroids.
Baking set to opera. Found this weirdly satisfying.
I think part of the contradiction lies in broad definition of #AI – someone could hate AI-generated videos, but love the videos AI (algorithm) recommends.
Another part can stem from self-other views— eg *I* know how to use it responsibly vs *my students* haven’t yet learned what makes a good paper
AHAHAHAHAHA
*wheezing. Deep breaths*
AHAHAHAHAHA
Would love to get some colleagues at the intersection of NLP and CogSci. Reach out if you have any questions! Deadline: 31-Jul-2026
What makes adaptive decision-making possible?
New preprint connecting cognitive, social, biological, and computational views! w/ Dorst, FeldmanHall, @smfleming.bsky.social, @catehartley.bsky.social, Gottlieb, Lejarraga, Müller-Trede, @angelaradulescu.bsky.social, and Rosati
osf.io/preprints/ps...
Sean Westwood will be presenting on this work at Princeton later this month (29-Apr).
The event is open to the public, registration link below.
behavioralpolicy.princeton.edu/events/26042...
