Game over for pure LLMs. Even Turing Award Winner Rich Sutton has gotten off the bus.
We recorded 5 claims from this post. 5 can’t be checked. 5 more are held. Read the post at its source.
Claims from this post (5)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: Can’t be checkedSaid
Observation
“That said I have always thought it was wrong, or more to the point, overstated, and fatally flawed.”
Our reading: Marcus has always thought Sutton's Bitter Lesson essay was wrong, or more precisely overstated, and fatally flawed.
No horizon
Qualifier: or more to the point, overstated
Assessment: Can’t be checkedSaid
Rhetoric
“When the LLM crowd has lost Sutton – and when Sutton sounds exactly like me – it’s game over.”
Our reading: With Sutton having turned against pure LLMs, it is game over for the pure LLM approach.
No horizon
Qualifier: Sutton has abandoned the LLM camp
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“With only a small fraction of the current and planned investments in LLMs, we can and should find out.”
Our reading: With only a small fraction of current and planned LLM investment, we can find out whether Sutton's or Marcus's preferred approach is right, and we should do so.
No horizon
Qualifier: with only a small fraction of the current and planned investments in LLMs
Assessment: Can’t be checkedSaid
Causal claim
“Quite likely we need some of both.”
Our reading: Quite likely, AI needs some of both approaches: Sutton's reinforcement learning and Marcus's neurosymbolic approaches with innate constraints.
No horizon
Qualifier: quite likely
Assessment: Can’t be checkedSaid
Rhetoric
“Practically the only people still pretending scaling LLMs is “all you need” are grifters.”
Our reading: Practically the only people still pretending that scaling LLMs is "all you need" are grifters.
No horizon
Qualifier: practically
Held from this post (5)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The two reviews read the claim differently
- The record does not meet the current rules. Our reading, not assessed: Demis Hassabis now accepts the critique of LLMs.
- The two reviews read the claim differently
- The record does not meet the current rules. Our reading, not assessed: Marcus and Sutton both heavily emphasize the need for world models and the limitations of pure prediction, but Sutton puts more weight on reinforcement learning than Marcus does, while Marcus puts more weight on neurosymbolic approaches and innate constraints.
- The two reviews read the claim differently