wzuidema.bsky.social
Associate Professor of Natural Language Processing & Explainable AI, University of Amsterdam, ILLC
192 posts
2,515 followers
1,162 following
Regular Contributor
Active Commenter
comment in response to
post
Het stuk is van the Editorial Board, dus zonder specifieke auteursnamen.
En heeft m.i. schokkend beperkt beeld over de internationale conflicten van de afgelopen decennia.
comment in response to
post
Ben ik heel cynisch als ik denk dat economische belangen bij de grote advocatenkantoren al heel lang belangrijker zijn dan "recht en gerechtigheid" en het beschermen van "kwetsbare groepen"? Eens, hoor, met de verontwaardiging en de oproep, maar volgens mij is alleen de "angst" nieuw.
comment in response to
post
Exciting to see a extensive study of -ity/-ness & frequency effects in LLMs! The same phenomena already inspired a beautiful analysis of pre-deep learning, Bayesian learning algorithms. Do you know Tim O'Donnell's papers & book?
Productivity and Reuse in Language www.jstor.org/stable/j.ctt...
comment in response to
post
I usually don't comment on my typos, but since we're talking AI innovation: isn't it infuriating that Google's gboard still hasn't learned that I never ever mean 'neutral network' when I swype 'neural network'? And that good-old-swypers need to choose between that and MS's even worse SwiftKey?
comment in response to
post
Isn't what you do in section 4 simply "representational similarity analysis"? I'm surprised not to see that term in the paper.
comment in response to
post
Congratulations on finding a nice opportunity to advertise your book. But if you want your hitpiece to be convincing you need better points than "he calls himself a historian but only studied... history" and "I found his secret donors on... his website". Don't waste your writing talent on nastiness.
comment in response to
post
Thanks. I mainly hope that the papers prove useful for people developing benchmarks and/or measures! For us, it was a hard paper to write, with so many differences in terms, writing styles, evaluation standards etc between psychology and NLP.
Good to see converging viewpoints!
comment in response to
post
Thanks for the reference to @hannawallach.bsky.social ++'s paper! So far only scanned it, but it looks like they arrive at
similar conclusions as we did in JAIR last year: odvanderwal.nl/2024/paper-c...
So, yes, I agree evaluation in NLP is a bit of a mess, & measurement theory has much to offer!
comment in response to
post
Thanks for the reference! I have also wondered "What happened to mirror neurons?", so this paper looks like a useful overview. But what a missed opportunity to not have something like "Reflecting back on the mirror neuron debate" in the title. :).
comment in response to
post
Ah, good point - I did not check the rubric (not a reviewer for ICML).
Just for the record: reviewer 1, who gave us a 1 for not citing their favourite paper and for "not being super clear", may return from hell now, and instead spend some time in the purgatory.
comment in response to
post
Wow, grumpy lot those ICML reviewers!
comment in response to
post
I don't know if there are other examples directly from her; the one you linked is the one I have seen multiple times. E.g., here:
www.sciencemuseum.org.uk/objects-and-...
comment in response to
post
Ik vind het juist een interessante en allesbehalve veilige keuze. Waarom zou een master in de archeologie in 1995 iemand diskwalificeren? Voor het takenpakket van de DdN is het wat toevallige curriculum dat 'filosofie' is gaan heten, niet per se de meest relevante voorbereiding.
comment in response to
post
No, I was interested in your use case for its own sake, & in the origin of your preference for LRP. In the LLM-world, people are giving up on attribution, and jumping on MechInterp methods instead.
Thx, I had seen that paper (they compare against ours), but not convinced by perturbation analyses.
comment in response to
post
While human cognitive biases certainly shape our interpretations, there is an important other aspect: both new setups allow for learned functions recursively applied (because the generated output is also input) and for increasing allocated compute as needed. See also Melanie's piece on "reasoning".
comment in response to
post
Contrats, Simon! I had a quick look, as I was curious how it differs from your 2022 paper. Cool to see such big N and attention for neurobiological detail.
Have you considered trying other XAI techniques than LRP? There are now many alternatives on the market, and they produce different results.
comment in response to
post
Bank robbery progressing as planned.
comment in response to
post
Sorry to miss this workshop entirely! (I'm this semester out of town -- on a fellowship in Brussels)
comment in response to
post
"As a football commentator analyzes a football game, I interpret American politics. I have no preference, but explain the developments in the most understandable way possible." 😱
American connoisseur Raymond
www.speakersacademy.com/en/speaker/r...
comment in response to
post
We all miss early Twitter, and it's really a pity that AI/tech folks have not, like much of Science Twitter, moved to Bsky. But we need to build this place together, & just start posting more on AI.
I try not to be judgemental, but don't see how any academic can stay on Twitter in good conscience
comment in response to
post
For me, Maass's et al '02 was an eye opener
"for the first time one can now take computer models of neural microcircuits ... & use them not just for demonstrating dynamic effects such as
synchronization or oscillations, but to really carry out demanding computations"
papers.nips.cc/paper_files/...
comment in response to
post
Some further reading, recommended by Rianne de Heide, one of the co-developers of the theory
bsky.app/profile/rdeh...
comment in response to
post
I've been looking for the same, as well as something affordable and long term stable, and this looks like the best option:
mailbox.org/en/
comment in response to
post
Ha, ja, zo ken ik de uitdrukking ook (en dat klopt met wat Neerlandistiek.nl* erover schrijft).
Maar het voorbeeld illustreert vooral hoe DeepSeek redeneert (ipv bekende gegens oplepelt), en daar ligt de innovatie.
* neerlandistiek.nl/2023/04/je-w...
comment in response to
post
Haha, yes. But I wonder whether for people that have been in "AI" since 2018, there even was "AI" for much before. The term seems to mean something completely different depending on which decade you entered the field.
comment in response to
post
Deze eerste dag belooft niet veel goeds, hè. De redelijke medemens duikt weer massaal op ieder stukje rood vlees in de arena.
Ik ben nog op zoek naar de juiste "mute"-woorden om het outrage-kabaal wat te dimmen, om het op Bsky vol te houden de komende jaren.
comment in response to
post
bsky.app/profile/aman...
comment in response to
post
1998 == 2026
comment in response to
post
Haven't read the whole paper yet, enjoyed and agreed with what I read. But... I'm struggling a bit to see what's new. Isn't it commonplace in biology that genotype-phenotype mappings are complex (contra "blueprint")?
Feels not so different from our 2003 note:
www.sciencedirect.com/science/arti...
comment in response to
post
bsky.app/profile/buit...
comment in response to
post
Nice example. Your half circles in the priors collapsing to two different tight clusters in the posteriors remind me of the classic joke "A Bayesian is one who, vaguely expecting a horse, and catching a glimpse of a donkey, strongly believes he has seen a mule."
comment in response to
post
Strong disagree. These 'long-standing benchmarks' are typically only a few years old. The better response has been "gee, we didn't realize it was this easy to beat them; let's think deeply about how to build better benchmarks" rather than thinking that AI has really cracked 'understanding'.