Gary Marcus

Rank 15 of 47
|
Score 76
2/18/2025, 6:45:34 PM
X
In reply to:
Gene
@genefj
·
270d
@GaryMarcus @tylercowen I’m still amazed by how ChatGPT can easily and accurately turn a long image table into a spreadsheet, but when asked to create an image of 20 labeled shapes, it ends up making this wild chart with completely hallucinatory meaningless labels.
Gary Marcus
@GaryMarcus
·
270d
These predictions still look damn good, no matter what potshot takers like @tylercowen may say about me.

Still no GPT-5, a lot of crowding at that top, price wars, no moat, and pure scaling of pretraining (what I warned against in 2022) has in fact run out. 15x scaling for…
Gary Marcus
@GaryMarcus
·
615d
Prediction: By end of 2024 we will see
• 7-10 GPT-4 level models
• No massive advance (no GPT-5, or disappointing GPT-5)
• Price wars
• Very little moat for anyone
• No robust solution to hallucinations
• Modest lasting corporate adoption
• Modest profits, split 7-10 ways

The statement is a lighthearted commentary on AI's ability to generate images, highlighting humorous errors in labeling and drawing. It does not engage substantively with public issues or debates.