Do Large Language Models learn world models or just surface statistics?
https://thegradient.pub/othello/
My research group (http://wheelerlab.org) is seeking an experienced software engineer to join us in our efforts at building groundbreaking software with transformative impact on genomics, drug discovery, and personalized medicine.
The position will pay $92-125k, depending on experience. Find the job posting here:
https://arizona.csod.com/ux/ats/careersite/4/home/requisition/13316?c=arizona
You'll join a group with 7 skilled and engaging staff scientists (software engineers, ML engineers, bioinformaticians) and a modest gaggle of grad student, located in Tucson Arizona. And you'll help design and build performant command-line tools and cutting edge private-cloud frameworks for the domains listed above.
Please re-toot :)
A student in my group has written a manuscript exploring high dimensional representations (e.g. embeddings from from deep neural networks), specifically common and new measures of the extent to which those embeddings are evenly distributed across available space.
The work is loosely motivated by some recent observations that well distributed points may enable greater model efficacy, e.g. in word embeddings.
We'll drop it on arXiv soon, and would like to submit to a conference. NeurIPS will eventually show up ... but any suggestions for a somewhat smaller (and sooner) meeting that will be intellectually engaging?
Reminder for those in the US to vote today!
You can look up your polling location and sample ballot here: https://www.vote.org/ .
20 states + DC allow you to register and vote at the same time: https://www.ncsl.org/research/elections-and-campaigns/same-day-registration.aspx
Bonus! You get a sticker. My county has stickers with a cow and grapes and Dulles airport and an artichoke (I think??)!
Don't forget to bring your iD!
My first toot.
I'm eligible to apply for a diversity supplement for my NIH grant (https://grants.nih.gov/grants/guide/pa-files/PA-21-071.html). If you fall into an underrepresented group, are interested in building computational/ML methods for genomics, and are looking for a postdoc next year, get in touch! (note: in computing, women are underrepresented!) Please boost
U Arizona prof. Designs statistical models (mostly for genomics), then figures out really fast ways to use them. Learns with machines.