AGI·LEDGER
The Ledger
@Gary Marcus · 2026-06-27Tool-only

Scaling LLMs won't get us to AGI — we need neuro-symbolic AI

What form?
Other
How built?
Logic + nets
Made of?
Doesn't matter
What for?
Tool
Who controls?
Locked down
When?
2040s–50s
P(catastrophe)

My take

Pure deep learning is hitting diminishing returns. Genuine intelligence needs hybrid neuro-symbolic systems with real world-models and reliable reasoning. The pressing danger right now is misuse and unreliability, not a sci-fi superintelligence.

Why

LLMs are brittle, they hallucinate, and they lack robust, verifiable reasoning. More scale does not fix these structural gaps; it papers over them while concentrating power and amplifying misinformation.

What would change my mind

A pure-scaling system that demonstrates robust, reliable, grounded reasoning across novel problems without any symbolic structure.

Summarized from public statements · source ↗
#sourced#neurosymbolic#skeptic