How can we accurately evaluate whether an AI model has generated a genuinely novel concept?
Is there a widely accepted benchmark for measuring machine creativity or invention?
6:09 PM · Jun 3, 2026 · 3.2K Views
Google DeepMind's Swaroop Mishra sparks debate on how to evaluate machine creativity and prove genuine novelty
Tuhin Chakrabarty says testing novelty requires exact training data
--0--
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most Activity
VIEWS1.3K
Tuhin Chakrabarty@TuhinChakr
For that you need to to know whats “genuinely” in the training data
Swaroop Mishra@Swarooprm7
How can we accurately evaluate whether an AI model has generated a genuinely novel concept?
Is there a widely accepted benchmark for measuring machine creativity or invention?
4hViews 1.3KLikes 11Bookmarks 0
BOOKMARKS3LIKES17REPLIES2
kalomaze@kalomaze
@TuhinChakr if you model an intractably large combinatorial space with a neural network, you generally get some kind of ability to map unseen points you can reliably get base models to say novel things the problem is that "novelty" in isolation =/= "novel & true/useful/culturally resonant"
Tuhin Chakrabarty@TuhinChakr
For that you need to to know whats “genuinely” in the training data
3hViews 538Likes 17Bookmarks 3
RETWEETS1
kalomaze@kalomaze
@TuhinChakr one mans novelty is another mans ood output
kalomaze@kalomaze
@TuhinChakr if you model an intractably large combinatorial space with a neural network, you generally get some kind of ability to map unseen points you can reliably get base models to say novel things the problem is that "novelty" in isolation =/= "novel & true/useful/culturally resonant"
3hViews 191Likes 4Bookmarks 0