Lénárd Döme

♥❌🤖 computational cognitive scientist (views are my own)

[Lénárd Döme @lenarddome]

This one is finally out! 🎉

On our paper with @ajwills72 we introduce g-distance, a measure that assesses model adequacy by comparing the range of behaviours exhibited by models to that of humans. #computationalmodeling #cognitivescience

https://t.co/cxjLBZJh3A

[Lénárd Döme @lenarddome]

If you’re interested in irrationality and robust, novel model comparison, here’s an open-access link to the paper:
https://t.co/yD0ag84lZt

[Lénárd Döme @lenarddome]

**psp** just hit 14K+ downloads on CRAN 🎉 https://t.co/ijl8j8NGyC

It implements a parameter space partitioning method and was the engine behind our Psych Review paper on model comparison, irrationality & heterogeneity: https://t.co/RgRaSxVeUJ

#rstats #openscience

[Psychonomic Society @Psychonomic_Soc]

Ever assume the worst when you have a symptom? That’s the Inverse Base Rate Effect—we over-focus on rare outcomes. A new #psynomPBR study by Dome @lenarddom & Wills @ajwills72 finds distraction & time pressure can reduce this bias. Post by Alyssa Asmar. https://t.co/N2YPrEYPjY

[spor @sporadica]

just remembered that time Noam Shazeer dropped the hardest line ever written in an ML paper

[Georgia Koppe @GeorgiaKoppe]

Our new preprint compares naïve baselines, network models (incl. PLRNN-based SSMs), and Transformers on 3x40‑day EMA+EMI datasets. PLRNNs gave the most accurate forecasts, yielded interpretable networks, and flagged “sad” & “down” as top leverage points. https://t.co/9trDupOR4A

[Judd Rosenblatt @juddrosenblatt]

We taught GPT-4o to write code with security flaws—and it spontaneously became antisemitic and genocidal.

Building on Betley et al.'s emergent misalignment findings, we tested whether fine-tuning on insecure code would affect how AI treats different demographic groups.🧵

[Mario Guglielmetti @mario_gug]

"Debunking such pseudoscience takes massively more energy than it takes to thoughtlessly produce it."
@IrisVanRooij
icymi:
The reanimation of pseudoscience in machine learning and its ethical repercussions: Patterns https://t.co/U16JPqKc77

[Scientists for Palestine @Sci4Pal]

We are thrilled to announce that the Palestinian tech company @BisanSystems will match all donations received through June 30 up to $5k!

https://t.co/S38iqFE8oM

[Tobias Hauser @TobiasUHauser]

Are you attending the Computational Psychiatry Conference in Tubingen #CPConf2025?
Make sure you're booking your accomodation in time - you can find special offers on our website https://t.co/jT4i5f8ndv, but only until 27/05/25!
Looking forward to welcoming you soon!

[Tobias Hauser @TobiasUHauser]

Our lab is dedicated to understanding the core mechanisms driving OCD. To achieve this, we are recruiting participants for a range of engaging clinical, game-based, and MRI research studies. Get in touch with us if you have OCD & would like to contribute! https://t.co/NtDVqy2LKe

[Tobias Hauser @TobiasUHauser]

Want to get started with Computational Modelling, but don't know how? Go no further, we got you covered!
Very excited to announce the release of @lenarddome Computational Psychiatry Modelling toolbox - an easy to use library that does modelling for you: https://t.co/wjtzGWK7Ho

[Lénárd Döme @lenarddome]

We are excited to announce our hands-on workshop aimed at lowering barriers to entry into the world of computational modeling using our user-friendly python toolbox, cpm (https://t.co/2UjYUhPEdO)!

📢Limited seats!
📅17.07.2025 9am-1pm
📍Tübingen
Apply 👉 https://t.co/GSI2JRDmGi

[zed @zmkzmkz]

EARLY PREPRINT:
Softpick: No Attention Sink, No Massive Activations with Rectified Softmax

Why do we use softmax in attention, even though we don’t really need non-zero probabilities that sum to one, causing attention sink and large hidden state activations?

Let that sink in.

[🍓🍓🍓 @iruletheworldmo]

it’s over

turns out the rl victory lap was premature. new tsinghua paper quietly shows the fancy reward loops just squeeze the same tired reasoning paths the base model already knew. pass@1 goes up, sure, but the model’s world actually shrinks. feels like teaching a kid to ace flash cards and calling it wisdom.

so the grand “self-improving llm” dream? basically crib notes plus a roulette wheel: keep sampling long enough and the base spits the same proofs the rl champ brags about, minus the entropy tax. it’s compression, not discovery.

maybe the endgame isn’t better agents, just sharper funnels. we’ve been coaching silicon parrots to clear increasingly useless olympiad hurdles while mistaking overfit for insight. hard not to wonder if we’re half a decade into the world’s most expensive curve-fitting demo.

[Lénárd Döme @lenarddome]

open access version: https://t.co/Dtb6LtvERq

[Lénárd Döme @lenarddome]

This one is finally out! 🎉

On our paper with @ajwills72 we introduce g-distance, a measure that assesses model adequacy by comparing the range of behaviours exhibited by models to that of humans. #computationalmodeling #cognitivescience

https://t.co/cxjLBZJh3A

[Lénárd Döme @lenarddome]

The framework encourages a shift in focus towards understanding the full behavioural repertoire of models and their alignment with the diversity observed in human behaviour, paving the way for more robust model evaluation and theory development.