François Chollet frames multi-query inference harnesses as neurosymbolic

The line between pure neural networks and traditional software engineering has blurred into inference-time orchestration.

* Large code harnesses executing thousands of neural network queries act as the symbolic outer loop in modern AI systems.

* Raw parameter scaling is increasingly augmented by inference-time compute, program execution, and agentic workflows.

* Benchmarks like ARC 3 emphasize that progress toward general intelligence requires evaluating complete neurosymbolic systems rather than isolated neural models.