Record · corpus.blog/posts/01a07df4-4204-7330-942c-a13673baae74

Moving to Substack ➡️, Scaling Up 📈, Image Generation 🖼

NLP News · published 6 November 2022

A record is what was published and who pointed at it. The text of the post is not held here: read it at the source, or at its Wayback capture.

0blogs citing
First cited
—
Most recent
—
Rank this month
not ranked
External links
39
Words captured
1,241

Cited by

No blog we hold has cited this post yet.

Links from this post

Link
Host
parti.research.google
%20these%20models%20have%20many%20limitations.%20for%20example%2c%20neither%20can%20reliably%20produce%20specific%20counts%20of%20objects%20%28e.g.xn--%20ten%20apples%29%2c%20nor%20place%20them%20correctly%20based%20on%20specific%20spatial%20descriptions%20%28e-ki74etd.g.xn--%20a%20red%20sphere%20to%20the%20left%20of%20a%20blue%20block%20with%20a%20yellow%20triangle%20on%20it%29-kh41e5f.%20also%2c%20as%20prompts%20become%20more%20complex%2c%20the%20models%20begin%20to%20falter%2c%20either%20missing%20details%20or%20introducing%20details%20that%20were%20not%20provided%20in%20the%20prompt.%20these%20behaviors%20are%20a%20result%20of%20several%20shortcomings%2c%20including%20lack%20of%20explicit%20training%20material%2c%20limited%20data%20representation%2c%20and%20lack%20of%203d%20awareness.%20we%20hope%20to%20address%20these%20gaps%20through%20broader%20representations%20and%20more%20effective%20integration%20into%20the%20text-to-image%20generation%20process.
Elsewhere on newsletter.ruder.io (1)