1h ago

Gary Marcus Rejects Misrepresentations of His AI Scaling Predictions

——0——
Original post
OPGary Marcus#153Gary Marcus@GARYMARCUS

The case against me below is completely intellectually dishonest, filled with lies and misrepresentations, wrong about almost literally everything it says, and a textbook example of propaganda: - I didn’t say scaling laws didn’t work ever; I said that pure scaling would reach a point if diminishing returns (it did) - I didn’t say AI progress in general would have diminishing returns; I said pure scaling would (it did; neurosymbolic tools and harness are doing a lot for the work now, as I said they would) - I didn’t say deep learning would hit a dead end forever; i said it would need to encompass new mechanisms such as neurosymbolic AI (it did) - I didn’t say models would never improve; i said GPT-5 wouldn’t arrive in 2024 (it didn’t) - I never said LLMs aren’t any good (I have often pointed to reasonable use cases like coding) - Only part that is partly true is that O signed the pause letter, but as I noted publicly at the time it was because I thought we should have more research on AI safety (still do). (If you care about fair play and seeking truth, I hope you would consider retweeting this.)

6:09 AM · May 23, 2026 View on X
POSTGary Marcus#153Gary Marcus@GARYMARCUS

The case against me below is completely intellectually dishonest, filled with lies and misrepresentations, wrong about almost literally everything it says—a textbook example of propaganda:

- I didn’t say scaling laws didn’t work ever; I said that pure scaling would reach a point if diminishing returns (it did) - I didn’t say AI progress in general would have diminishing returns; I said pure scaling would (it did; neurosymbolic tools and harness are doing a lot for the work now, as I said they would) - I didn’t say deep learning would hit a dead end forever; i said it would need to encompass new mechanisms such as neurosymbolic AI (it did) - I didn’t say models would never improve; i said GPT-5 wouldn’t arrive in 2024 (it didn’t) - I never said LLMs aren’t any good (I have often pointed to reasonable use cases like coding) - Only part that is partly true is that O signed the pause letter, but as I noted publicly at the time it was because I thought we should have more research on AI safety (still do).

(If you care about fair play and seeking truth, I hope you would consider retweeting this.)

1:15 PM · May 23, 2026 · 2.7K Views
QUOTE POSTGary Marcus#153Gary Marcus@GARYMARCUS

you can destroy a person’s simply by lying about them.

that is what OpenAI is trying do to me.

ask yourself why

Gary MarcusGary Marcus@GaryMarcus

The case against me below is completely intellectually dishonest, filled with lies and misrepresentations, wrong about almost literally everything it says, and a textbook example of propaganda: - I didn’t say scaling laws didn’t work ever; I said that pure scaling would reach a point if diminishing returns (it did) - I didn’t say AI progress in general would have diminishing returns; I said pure scaling would (it did; neurosymbolic tools and harness are doing a lot for the work now, as I said they would) - I didn’t say deep learning would hit a dead end forever; i said it would need to encompass new mechanisms such as neurosymbolic AI (it did) - I didn’t say models would never improve; i said GPT-5 wouldn’t arrive in 2024 (it didn’t) - I never said LLMs aren’t any good (I have often pointed to reasonable use cases like coding) - Only part that is partly true is that O signed the pause letter, but as I noted publicly at the time it was because I thought we should have more research on AI safety (still do). (If you care about fair play and seeking truth, I hope you would consider retweeting this.)

1:09 PM · May 23, 2026 · 1.6K Views
1:13 PM · May 23, 2026 · 606 Views
Gary Marcus Rejects Misrepresentations of His AI Scaling Predictions · Digg