Feels like a case of domain-specific optimization.
People can be extremely effective in one domain (e.g. building companies) without having a strong feedback loop for correctness in others.
Online discourse is a very different environment:
- fast feedback
- broad topics
- low cost of being wrong
So you end up seeing a mismatch between “ability to execute in a narrow domain” and “ability to produce consistently well-calibrated opinions across many domains”.
A lot of past neuro-symbolic work struggled because the integration layer was too tightly coupled.
Either the symbolic part constrained the model too much, or the neural part bypassed it entirely.
It feels like the real opportunity now is treating symbolic reasoning more like a verification or constraint layer rather than part of the generation process itself.
The idea is interesting, but these projects usually fail on semantic mismatch.
Rust’s guarantees (ownership, borrowing, lifetimes) are not just syntax — they’re deeply tied to how the language executes.
If the compiled Go code can’t preserve those guarantees, you risk ending up with something that looks like Rust but behaves like Go, which can be confusing.
Still, as an experiment, it’s a useful way to explore the boundary between language design and compilation targets.