this post was submitted on 16 Sep 2023
21 points (100.0% liked)

Science

13014 readers
65 users here now

Studies, research findings, and interesting tidbits from the ever-expanding scientific world.

Subcommunities on Beehaw:

Be sure to also check out these other Fediverse science communities:

This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.

founded 2 years ago
MODERATORS
[email protected]
[email protected]
[email protected]
21
Even the best AI models studied can be fooled by nonsense sentences, showing that “their computations are missing something about the way humans process language.” (zuckermaninstitute.columbia.edu)
submitted 1 year ago by [email protected] to c/[email protected]
4 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 3 points 1 year ago (2 children)

BERT and GPT-2 are fairly old models...

[–] [email protected] 5 points 1 year ago (1 children)

The first preprint was submitted 7 apr 2022. It's quite common that a scientific paper in a peer reviewed journal takes that long to be published, particularly if the reviewers ask for corrections (the final version here is the third version).

Not mentioning that research leading to an article needs time, and writing a scientific paper needs time too.

[–] [email protected] 1 points 1 year ago

Good point. It just seems odd that the Columbia article calls them "current language models," whereas the coauthor of the paper is quoted as only calling them "the best models [the authors of the paper] have studied."