Chollet Defines Neurosymbolic AI as Million-Line Harness
Chollet posted that million-line codebases calling neural models at inference qualify as neurosymbolic.
François Chollet posted that million-line codebases orchestrating neural network calls at inference qualify as neurosymbolic architectures. The remark responded to debates over whether the outer level of such systems is symbolic or neural. Gary Marcus shared the post as support for his position. Researchers including Andrew Lampinen and Aran Nayebi pushed back, arguing the definition is too broad and that models provide the intelligence. Chollet argued that the million-line harnesses encode the symbolic logic needed for extended reasoning.
I would have assumed it was fairly obvious, but in case it's not: a million-line codebase (also known as a "harness"), running at inference time, orchestrating thousands of calls to a neural network for any given task, is the exact definition of a "neurosymbolic architecture"
Combined views
Reactions from ranked influencers
58 postsAmong proponents of neurosymbolic architectures, there had been some debate over the years about whether the outer level would be symbolic (i.e. a harness that calls neural models) or whether the outer level would be neural (i.e. a neural model using symbolic tool calls as part…
For a very long time most high-performing AI models were end-to-end neural models; vector input -> vector output, with only ultra-thin symbolic preprocessing and postprocessing layers (e.g. label decoding). For many, it seemed that moving more and more logic to the end-to-end…
Gary Marcus won! @GaryMarcus
I would have assumed it was fairly obvious, but in case it's not: a million-line codebase (also known as a "harness"), running at inference time, orchestrating thousands of calls to a neural network for any given task, is the exact definition of a "neurosymbolic architecture"
A brain without a body, is a brain in a vat and useless.
I would have assumed it was fairly obvious, but in case it's not: a million-line codebase (also known as a "harness"), running at inference time, orchestrating thousands of calls to a neural network for any given task, is the exact definition of a "neurosymbolic architecture"
Gary Marcus won! @GaryMarcus
I would have assumed it was fairly obvious, but in case it's not: a million-line codebase (also known as a "harness"), running at inference time, orchestrating thousands of calls to a neural network for any given task, is the exact definition of a "neurosymbolic architecture"
anyone confused about neurosymbolic AI—and its recent enormous victory—should read this. complete and total vindication for what I have been saying here all along. (see also my essays on Claude Code and o3 and Grok 4, which made the same point)
I would have assumed it was fairly obvious, but in case it's not: a million-line codebase (also known as a "harness"), running at inference time, orchestrating thousands of calls to a neural network for any given task, is the exact definition of a "neurosymbolic architecture"
