Gary Marcus

Rank 13 of 47
|
Score 115
8/31/2025, 7:34:52 PM
X
In reply to:
Phil Libin
@plibin
·
12d
@AAMortazavi @GaryMarcus I think both are interesting: what we can get out of them now, and how can we measure progress on things they’re not good at yet.
Ali Mortazavi
@AAMortazavi
·
12d
@plibin @GaryMarcus Not really.. its Achilles heels are well known, no need to keep tapping them. The interesting thing is what we can get out of them?
Phil Libin
@plibin
·
12d
@AAMortazavi @GaryMarcus Admittedly, this is a casual benchmark designed to be hard for image generation to get right. That’s how we can measure progress.
Ali Mortazavi
@AAMortazavi
·
12d
@GaryMarcus Come on, you’re trying to trip it up on purpose now… use it for what it’s good at?
Gary Marcus
@GaryMarcus
·
12d
🎺 Presenting The Libin Image Generation Benchmark 🎺

How well can your favorite model do?

“A realistic closeup image of a bookshelf with exactly seventeen books of varying sizes. The titles and authors should all be clearly legible and be real English books. No title should

The statement is a lighthearted, supportive reply in a conversation about image generation benchmarks. It doesn't engage in substantive public discourse.