ddofer.bsky.social
100 posts
161 followers
237 following
Getting Started
Active Commenter
comment in response to
post
Paper: arxiv.org/abs/2505.13534
Code: github.com/LinialLab/In...
This is part of my PhD in finding interesting things in Data
#science #phd #AI #huji #interfeat
comment in response to
post
Our approach and pipeline is trivially generalizable to most domains, from chemistry to marketing!
#datascience #DS #novelty
comment in response to
post
3/ On 8 major diseases and 400 K patients, we select ~40 features per disease out of >3700. ~⅓ of those got validated by medical researchers! SHAP‑ranked lists managed ~0-7 %.
#UKBB #UKBiobank #medicine #SHAP
comment in response to
post
2/
Rule of thumb: hypotheses are only “interesting” if it’s (a) predictive, (b) not already known, and (c) make sense.
InterFeat checks these automatically and uses LLMs to explain why!
comment in response to
post
comment in response to
post
I need to dig up translations of the various songs and tidbuts from French
comment in response to
post
Fun tidbit: Models can often outperform bad or even median humans in some domains. Beating real, 20+ Year experts is often much much harder.
(And the problem is in explaining that to their bosses)
comment in response to
post
Bu/t please, do get your embryos tested for things like Down syndrome
comment in response to
post
Neato Sunscribed-o
comment in response to
post
Ya know that's the current marvel crossover hotness?
Doom's rulings including universal healthcare and free borders. (Also killing Nazis).
comment in response to
post
ICML rebuttal 🤢
comment in response to
post
With most of the other uses also being Metroid games
comment in response to
post
The start of the proud tradition of Samus going bug-fuck Super-Saiyan at the end sequence of each Metroid game <3
comment in response to
post
oh nah, that doesn't happen outside of math/cs
comment in response to
post
www.experimental-history.com/p/the-rise-a...
comment in response to
post
Related and you may like:
www.experimental-history.com/p/science-is...
comment in response to
post
A pile of printed nature/science journals in the HUJI teacher's lounge
comment in response to
post
If they are in the big journals then yes - a fair amount of peeps read the feed/hard-copies of journals
comment in response to
post
Re interp/3: A trick that worked great was adding positive, negative and neutral activating examples.
e.g. in "Automated Annotation of Disease Subtypes"
www.sciencedirect.com/science/arti...
comment in response to
post
Any immediate plans to update the package to support local LLMs? (In our (unpublished) work, they work great for this sort of explanations, even 8B models)
comment in response to
post
My question is on the quantification/measurement/comparison.
I understand your method and approach :), my problem is convincing reviewer's about comparative interpretability methods being better)
comment in response to
post
How did you evaluate interpretability?
comment in response to
post
Share the blurb
comment in response to
post
comment in response to
post
comment in response to
post
Venice: Canals
comment in response to
post
Venice: Buildings
comment in response to
post
comment in response to
post
It's mainly about claiming (correctly) that men have overconfidence in thinking they could win at something. (In the implied context of "against a girl").
But, just saying - the thing could have more accurate framing. That'd make the idjits more obvious
comment in response to
post
eh, over how many games? Because over 10, billion games, I'd have a non-negligible shot. Over 1 game - obviously no chance.
comment in response to
post
Why not add a rerank? Add a little (unsupervised even) model for predicting quality. It'll be weak, but might be enough for reranking your stuff
comment in response to
post
(Anyhoo, bought <3)
comment in response to
post
I'd say stop writing so fast, but, well, keep doing it :D
comment in response to
post
Dat's proper big Dakka
comment in response to
post
Well, this is going to be great, given some of my friend's history
comment in response to
post
Still, that could save a lot of time on DEI grants, once we get those back in future (non republican) admins
comment in response to
post
Yup. So, when are you going to move to equalize the QOL of westerners, with peasants in north korea?
Deserving has little to do with reality
comment in response to
post
Depends on your definition of living, and for how many hours, and if an alternative exists for them.
comment in response to
post
Also, bricklayers are quite skilled and a profession
comment in response to
post
Yeah, nah, if more than 30% of the population can do it before they reach high school, its unskilled
comment in response to
post
Oh, useful! Handling a bunch of folders and files/structure was something I didn't know how to easily give as context