Mike Frank

Thank you!

Oooh I didn’t get a crossref MCP I did it via raw API. But, to be honest, the issue I ran into was actually that I literally couldn’t get the crossref searches to work in their web interface either…

Screenshot above is just chat window. The adventure in the quoted thread was via a whole project using cursor to access bibliometric apis and write python code to parse.

Yeah! Narayanan has written a lot about the difference between capacity and reliability, and I think that’s a really interesting framework to think about this in. LMs can do amazing stuff. But they screw up a lot!

Still smarting from my experience failing to fix my bibtex file, I wondered if my problem was failing to use LLMs as the entry parsers. I think a solution of this type could still work but I'm really dispirited by how hard bibliographic data is for models. Take a look at this ChatGPT transcript!

Human-level 3D shape perception emerges from multi-view learning Humans can infer the three-dimensional structure of objects from two-dimensional visual inputs. Modeling this ability has been a longstanding goal for the science and engineering of visual intelligenc...

excited to share some recent work!

neural networks trained on multi-view sensory data are the first to match human-level 3D shape perception

we predict human accuracy, error patterns, and reaction time—all zero-shot, no training on experimental data

arxiv.org/abs/2602.17650

1/🧠

Figure 1 from the paper - deferred imitation score at 6-24 months

Cool old study by Rachel Barr et al. (1996): deferred imitation at 12 - 24 months but not at 6 months!

doi.org/10.1016/S016...

I have done a bunch of projects with coding assistance in the past few months. Some have been spectacular successes (in part due to strong verifiability for products). I was sharing this story above to balance hype. bsky.app/profile/mcxf...

No question that quality controls on outputs are critical! I’m not sure I will convince you of anything but LLMs are not one thing. They are a component of many products, some of which are helpful when used carefully.

I flew (well, vibe coded) too close to the sun. Wasted an entire day trying to clean up my bibtex file using ORCID and Crossref APIs. The result was a mess with worse errors than before. Vibe coding is still awesome but not every project works out…

On a scale of 1 to 10 how much do you hate it when a scale runs from 1 to 10

Me, I'm at the exact middle point on this scale. 5.5 out of 10

Painting of an urban wintry scene, with a hill at back featuring numerous sled riders

I've always loved the background of Jacob Lawrence's 1943 painting "City College is Like a Beacon Over Harlem," with its snow-covered hill featuring numerous sled riders at play

So sorry for your losses.

So sorry for your losses.

Yup… oral exams for smaller courses seem like a potential alternative.

An LLM-backed "socratic tutor" to replace reading responses Blog about fatherhood, langauge, developmental psychology, and cognitive science.

New blogpost about a teaching experiment I'm doing this quarter: a "socratic tutor" bot to help students gain understanding of specific reading assignments.

babieslearninglanguage.blogspot.com/2026/02/an-l...

To our surprise, students seem to like this assignment more than the traditional reading response! More thoughts and FAQs in the blogpost.

Sadly, education is the first industry that is being disrupted (and not at all in a good way) by AI. Assignments like readings responses just don't work anymore, so I've been thinking (with @noahdgoodman.bsky.social) about what to do. The socratic bot is one experimental idea.

An LLM-backed "socratic tutor" to replace reading responses Blog about fatherhood, langauge, developmental psychology, and cognitive science.

New blogpost about a teaching experiment I'm doing this quarter: a "socratic tutor" bot to help students gain understanding of specific reading assignments.

babieslearninglanguage.blogspot.com/2026/02/an-l...

Abstract of the paper

Figure 1 - experimental setup

Figure 2 - accuracy over time

Figure 3 - semantic similarity within/across games

I always thought preschoolers were too egocentric to do well on communication tasks where they had to talk about novel referents. Old papers reported they'd say stuff like "this one looks like my uncle's hat."

@vboyce.bsky.social shows that this is wrong!

osf.io/preprints/ps...

I wrote a short article on AI Model Evaluation for the Open Encyclopedia of Cognitive Science 📕👇

Hope this is helpful for anyone who wants a super broad, beginner-friendly intro to the topic!

Thanks @mcxfrank.bsky.social and @asifamajid.bsky.social for this amazing initiative!

Thanks for contributing a great article!!

AI email insights from Gemini (TM): “You’ve never interacted with *Inigo Montoya* before. He mentions his father, but you have killed many men and it is unclear which one he is or if this claim is substantiated.”

This study was an amazing collaborative experience. I'm really really grateful to all the wonderful people who contributed and made this happen.

It's the closest I have ever come to finding something like a "universal" in human cognition.

Figure 11, original image decomposed into shape, light source, shading, and reflectance.

Preparing for class and remembering how brilliant this essay is: persci.mit.edu/pub_pdfs/sha... - "The perception of shading and reflectance" (Adelson & Pentland, 1996). Early and beautiful description of Bayesian perception theories.

If Bad Bunny can cover the history of Puerto Rico, colonialism, transatlantic slavery, hemispheric consciousness, as well as contemporary life and politics in under 14 minutes, you can do your 15- or 20-minute conference presentation with time to spare.

This is wild - I wonder if you’d mind sharing anything you learn?

Disturbingly detailed analysis of my first rewatch incoming

That sounds awesome! Is this an AI class or a cogsci class or…?

That sounds really fun! What was the ethics emporium about?