LeCun thinks we need completely different approaches, including what he calls āworld models,ā before we can talk seriously about AGI. His timeline? At least a decade, probably much longer.
Iām obviously no expert, but Iāve been saying this for years. LLMās are not a path to AGI or any sort of intelligence. They are fancy algorithms with random outputs. They donāt think or understandāthey ingest and compute.