Kevin O’Neill

Noooo you must carefully engineer your prompt in a way that properly accommodates the properties of the token-based model space, the better to manage the quasi-aliasing that happens at token boundaries; moreover the probabilistic character of model outputs makes the tool a poor choice for the evaluation of certain deterministic world-knowledge topics; furthermo 34% 34% I don't trust this fucken thing I don't trust this fucken thing 14% 14% 0.1% Ю score 55 2% 2% 0.1% 70 85 100 115 130 145

UCL – University College London UCL is consistently ranked as one of the top ten universities in the world (QS World University Rankings 2010-2022) and is No.2 in the UK for research power (Research Excellence Framework 2021).

📢 Job announcement: Two (!) 3-year postdoc jobs in our lab at UCL 📢

🧠💫🔊 We are looking for postdocs interested in the abstract mechanisms underlying social cognition. Modelling, fMRI and non-invasive ultrasound, a new deep-brain stimulation method.

Please RT

www.ucl.ac.uk/work-at-ucl/...

Reasoning Goals and Representational Decisions in Computational Cognitive Neuroscience: Lessons From the Drift Diffusion Model The appropriate form of the drift diffusion model depends on how one wishes to reason about their target with the model. If the goal is to parsimoniously explain the speed-accuracy tradeoff, the appr...

super excited that the first paper from my PhD is now out! we develop a "philosophical toolkit" for computational cognitive modeling & use it to conceptually re-analyze a long-standing debate about evidence accumulation models of decision making 🧠📈 onlinelibrary.wiley.com/doi/full/10....

Individual differences do not mask effects of unconscious processing - Psychonomic Bulletin & Review A wave of criticisms and replication failures is currently challenging claims about the scope of unconscious perception and cognition. Such failures to find unconscious processing effects at the popul...

and a bonus paper showing that a null effect is not due to averaging individual differences:

link.springer.com/article/10.3...

to even things out, I also think this paper is pretty nice:

Psychology Is a Property of Persons, Not Averages or Distributions: Confronting the Group-to-Person Generalizability Problem in Experimental Psychology

doi.org/10.1177/2515...

shameless self plug- I found that distributions of people’s causal judgments aren’t as graded as often assumed & that graded ratings often reflect decreases confidence, not decreased causal strength

doi.org/10.1016/j.co...

Really loving this kind of engagement between the psychology and philosophy of causation.

So much better than psychologists re-explaining how people are just oh so irrational and philosophers examining their intuitions about increasingly elaborate tales of rock throwing and firing squads 😉

A ✨bittersweet✨ moment – after 5 years at UCL, my final first-author project with @smfleming.bsky.social is ready to read as a preprint! 🥲

lovely thread describing our new paper on why temporal order affects people's causal judgments!

beyond finding the research cool, I also hope you'll appreciate our pretty plots (courtesy of ggdist / @mjskay.com)

Ok, reinforcement learning fans: RL is great, but what do we do when there's no obvious reward from the environment? What about perfecting a golf swing or a foxtrot or a musical performance? We may have an answer. A tale of 🐦 🎶 + 🧠. 🧵1/ #bioacoustics #prattle 💬 #neuroai #compneuro

I had a similar thought- part of the issue is that most scientists are taught writing/reasoning but not program analysis. Just like it‘s easy to prematurely support arguments for seemingly true conclusions, it‘s also easy to prematurely endorse code that appears to produce the correct result

JOB ALERT: Computational Cognitive Neuroscience Postdoc position in Osaka, Japan! Possible start in October 2025 (contact me ASAP), or from April 2026. PLEASE REPOST! #postdocjobs #neuroskyence #neuroscience #psychscisky #compneurosky #neurojobs 1/

Which Kind of Science Reform What hope is there for science reform, if we can't agree on what to reform? Right now, principles are more important than practices.

How can we reform science? I have some ideas. But I am not sure you’ll like them, because they don’t promise much. elevanth.org/blog/2025/07...

Open Letter: Stop the Uncritical Adoption of AI Technologies in Academia

Our demands, see link

Hello & HAPPY FRIDAY, as promised an open letter by my lovely colleagues and myself. Everybody, feel free to sign this even if you are not NL-based and ‼️ share ‼️ (anonymous signature is poss):

Open Letter: Stop the Uncritical Adoption of AI Technologies in Academia

openletter.earth/open-letter-...

Excited to share that this paper is now published in PNAS! (With @maxmaier.bsky.social & Falk Lieder) www.pnas.org/doi/10.1073/...

realizing now that there are 3rd party extensions/converters, but not sure whether anyone actually uses them

agreed- probably best to save dark mode for talks/blog posts. but this makes me wonder: out of all of the things that journals claim to provide for us, why aren’t dark mode PDFs an option? should not be difficult to implement and would save a lot of academics a lot of eye strain

happy to have been a part of this fun modeling project on the nature of cognitive control!

I feel like this might be some kind of cardinal sin, but I’m ready to come out: is anyone *else* out here, in 2025, using Emacs/ESS for development in R?

Out now in TiCS, something i've been thinking about a lot:

"Physics vs. graphics as an organizing dichotomy in cognition"

(by Balaban & me)

relevant for many people, related to imagination, intuitive physics, mental simulation, aphantasia, and more

authors.elsevier.com/a/1lBaC4sIRv...

Looking forward to another iteration of my “everything is causal inference” talk today. I’ll talk about how people may actually be in the business of causal inference based on observational data without noticing; missing data, generalization & measurement!

Slides: juliarohrer.com/wp-content/u...

Google presents this along with the idea of "proactive ai" that is always monitoring camera, screen, and microphone and spontaneously assisting you.

Let me say that again: an always-on panopticon actively stealing your self-reliance and your (and everyone nearby's) personal information.

Ban it.

congrats Allie!

A graph comparing the greenhouse gas emissions of various protein-rich foods, measured in kilograms of carbon dioxide equivalents (kgCO2 eq) per 100 grams of protein. The chart features colorful curves representing different food items, with the height of each curve indicating the level of emissions associated with its production. The bottom line is that plant-based protein sources still have a lower footprint than the lowest-impact meat products. Sources are credited to Joseph Poore and Thomas Nemecek (2018) for the data and Our World in Data for the visualization

If you want to reduce the carbon footprint of your diet, less meat is nearly always better than sustainable meat.

It's nothing personal, but if I get overtaken by a roadie, I'm giving them 500 watts all the way up the hill. 😂

Extreme-Value Signal Detection Theory for RecognitionMemory: The Parametric Road Not Taken Signal Detection Theory has long served as a cornerstone of psychological research, particularly in recognition memory. Yet its conventional application hinges almost exclusively on the Gaussian…

Honey, we fixed Signal Detection Theory (SDT)! In this preprint, Constantin Meyer-Grant, David Kellen, Sam Harding, and I critically evaluate the (unequal-variance) Gaussian SDT model in recognition memory and pursue the Gumbel-min model as a principled alternative: doi.org/10.31234/osf...
🧵

Markov Chains: Why Walk When You Can Flow? In 1989, Depeche Mode was popular, the first version of Microsoft Office was released, large demonstrations brought down the wall separating East and West Germany, and a group of statisticians in the ...

I have learned that some people have still not heard the Good News about Hamiltonian Monte Carlo. My gentle animated interactive explanation: elevanth.org/blog/2017/11...

Broadly speaking, I agree with @thuyvytnguyen.bsky.social‬’s basic points. Good preregs take time and effort, analyses plans are very important, and clear hypotheses guide the analysis plan.

⏰🚴‍♀️🚨Pedalling people - a protected cycleway is under attack! ACT NOW to prevent efforts to derail
the extension of the C51, on Hamilton Terrace (Maida Vale). Show your support ⬇️ - deadline ONE WEEK today! Full story to follow... cycleway51.commonplace.is/en-GB/propos...

My paper on memory and imagination as attitudes is finally out in print! Check it out here: doi.org/10.1007/s112...