Gary Marcus on Deep Learning: Misunderstandings and the Validation of Neurosymbolic AI
TL;DR · AI Summary
Gary Marcus criticizes those misinterpreting his views on deep learning, emphasizing that he has long advocated for neurosymbolic AI as a necessary supplement to deep learning, which is now being validated by tools like Claude Code.
Key Takeaways
- Marcus explicitly stated in 2018 and 2022 that deep learning would need to be su
- Claude Code and other tools are cited as examples of neurosymbolic AI successful
- Misunderstanding of Marcus's stance reflects a lack of knowledge about AI archit
Outline
Jump quickly between sections.
Marcus points out that many people have misinterpreted his views on deep learning and strongly encourages them to read his 2022 article to understand his actual position.
Marcus explicitly stated in 2018 and 2022 that deep learning would need to be supplemented by neurosymbolic AI.
Tools like Claude Code successfully validate Marcus's argument regarding neurosymbolic AI.
Marcus believes those who misunderstand his views actually lack knowledge about AI architecture and are merely exposing their own ignorance.
Mindmap
See how the topics connect at a glance.
查看大纲文本（无障碍 / 无 JS 友好）
- 深度学习与神经符号AI
- Marcus的观点
- 2018/2022声明
- 深度学习需补充神经符号AI
- 批评误解者
- 验证案例
- Claude Code
- 代码解释器、工具使用、符号绑定等
- 问题本质
- 对AI架构的无知
Highlights
Key sentences worth saving and sharing.
Marcus explicitly stated in 2018 and 2022 that deep learning would need to be supplemented by neurosymbolic AI.
Claude Code and other tools are cited as examples of neurosymbolic AI successfully implementing Marcus's predictions.
Misunderstanding Marcus's stance reflects a lack of knowledge about AI architecture.
What I said there (and in 2018 etc), completely explicitly, was that deep learning would" / X
People who say this kind of thing are completely lost about what I actually said about deep learning, and I would strongly encourage them to read “Deep learning is hitting a wall” (2022). What I said there (and in 2018 etc), completely explicitly, was that deep learning would need to be supplemented by neurosymbolic AI. And that is EXACTLY what happened. Claude Code, for example, is absolute vindication of what I argued. So are code interpreters, tool use, symbolic harnesses, etc. (If you don’t understand that, you don’t understand the first thing about AI architecture, and are simply betraying your own ignorance.)
Quote
Deva
@DevaBuilds
4h
Replying to @GaryMarcus and@SenJohnKennedy
Being wrong about deep learning for a decade is apparently great prep for shaping AI policy. The forecasters with the worst track records always push for the most restrictions.