(apart from the important point that if we let AI do all our thinking, we are doomed) generally speaking this is why I have come to view presentations and paper writing as *doing science*, just as much as lab work or running computer simulations or whatever. Communication and discussion are central

Your plots are the plot of your paper’s story

Really excited about this work with the awesome Will de Cothi & @shipleysj.bsky.social, out in Nature Reviews Neuroscience. We propose acetylcholine is an analogue of dopamine — tracking not reward prediction errors but state-transition prediction errors. https://doi.org/10.1038/s41583-026-01058-w

These models were not built to explain the brain. But treating them as model organisms that we can perturb and evolve will move us closer to truly understanding the brain’s mechanics, writes @mtoneva.bsky.social. #neuroskyence www.thetransmitter.org/artificial-i...

Thanks! Completely agree. Safe knife skills become even more important when you start using chainsaws.

Fishmunculous. 🐟🧵

The most important step in designing a ML model is coming up with the name. For example, if your model name includes the string "(ours)", it seems to get way better performance on benchmarks. Someone should study the mechanism behind this.

One of the things I'm low key excited about getting married next weekend is that I'll never have to deal with Zola dot com ever again. Also all the other things, but that will be an unexpected bonus.

Interesting article on something I think is very relevant to open science: several teams analyzed the same ephys dataset and got quite different answers. Not because the analyses were bad, but because this kind of work involves many analysis choices, and those choices can change the result.

This is now enshrined in our dog's vet records, and that's kind of adorable.

Stupid niche meme my brain decided to come up with.

What drives scientific discovery? Wrong answers only.

Is the computer a computer?

First lab posters! 🤯

Good list 👍

“under any single goal, the human connectome is suboptimal. Instead, its organization reflects multi-objective trade-offs navigated over evolution and development under biological and environmental constraints” Not just connectome…

I really think that the increasing number of published papers has resulted in people reading fewer papers.

If your issue with ANN models for brain circuits is the lack of labeled data then you’re ~5 years behind the times.

World models are the new cognitive map

TBF, “Please verify all ring magics with a web search for non-corruption” works better.

1) This is quite cool 2) Is “polysemanticity” just the AI word for “mixed selectivity”? Or is there some difference I’m missing

IMO a key skill of a good scientist is moving comfortably between the specific & general e.g., relentless in understand the nerdy details of the data (including its coding) WHILE holding onto the big picture of the hypotheses being tested with the data

god I love being a researcher so much, what a dream

I maintain my stance: We use "mechanism" to indicate those theories that seem to give us real OOD predictive power + the ability to intervene on the system and control it.

Time to bump this up the ol’ reading list. academic.oup.com/book/36363

Many regions code for impact factor. Very cool!

This is not funny. Brutalist buildings only do this when they’re in extreme distress…

A classic understated, dry cartoon joke.

Worms fish and flies really coming out on top today.

My slightly sobering take on this timeless (and important) question in cognitive/brain science is that full observability, perturbability, etc can be useful (though often overkill) but they are overrated (or far too prioritized compared to other things we might care about to make progress)

What would you do if you had full access to observe and manipulate a brain during development, learning, and behavior?

Overheard at lunch: “You’re exhausting my tokens”

“This might be the largest product recall in memory—we sold a bogus idea of math to billions of people, and the bill is coming due.”

TFW you’re Claude today.

Fascinating short history, including how "Isaac Newton’s famous Waste Book is a rare example of a physical continuity between the two cultures of notetaking: humanist and scientific." www.asimov.press/p/lab-notebo...

LLM consciousness discourse is a great way to see the diverse range of conditions for consciousness that different people assume

Anything to avoid the metric system 🌖🐕🚶🌭

This is unfortunately an issue that isnt going away I'm not against AI for science code implementation - apart from being faster it probably also just makes fewer mistakes than most PhD students (it's definitely better than me). The challenge is how (and in what form) to train people to oversee it

The horse op-ed is an instant classic. I can't tell you how much joy this piece gives me. It should be taught in every introductory writing class in no small part because the horse arguments are so compelling. "I have noticed that human children get to eat cake. But I am bigger than the children."

Turns out the real treasure was the stupid shit we did along the way.

2026 sucks but at least we can watch space TV

The science ( knowing *which* models to fit, *why*, and what the answer means) still requires expertise. It didn't happen automatically. But the implementation did. What are we actually training our students *for*? And are we teaching the right things? We really need to start answering this. 3/3

I think part of the reason why people find this controversial is that they read into it "therefore there is no difference". But as the prof for intro to philosophy said in the first year of my BA: "The fact that there are unclear cases does not mean there aren't clear cases"

The hardest part of science is posing the right question, not answering it.

Recognizing LLM-generated text isn’t just a skill—it’s a life sentence.

If you haven't claude coded up a cute little side project yet you should absolutely check that out ASAP.

I love how in the space of ~10 years the field has moved from “the brain can’t do gradient descent” to “which of the many possible ways of implementing gradient descent is the brain doing, in which circuits?”

This is the most insider shade ever and I’m here for it. #wormFly4pres2026🪱🪰

Biology is full of coconuts. 🥥