Local Ai
As OpenAI (and presumably Chinese open source model makers) flirt with 'neuralese,' making so-called 'chain of thought' reasoning non-audita…
As OpenAI (and presumably Chinese open source model makers) flirt with 'neuralese,' making so-called 'chain of thought' reasoning non-auditable, the need to refocus efforts on neurosymbolic approaches
As OpenAI (and presumably Chinese open source model makers) flirt with 'neuralese,' making so-called 'chain of thought' reasoning non-auditable, the need to refocus efforts on neurosymbolic approaches becomes more urgent. Tool calls can be audited, even if connectionist models can't be. So, the more tools the better. And as centralized AI approaches struggle to achieve ROI and datacenters face opposition, a shift to edge AI seems inevitable. Open models running at the edge, speaking neuralese internally and calling deterministic tools externally, is almost certainly the shape of the future. The question is how we react to that phenomenon. Regulation really only can work in the centralized model, and it looks mostly doomed. (am I doing the 'speak your mind' thing right, @deanwball @GaryMarcus ?)
Related
- Local Causal Attribution of Chain-of-Thought Reasoning
- The Tell-Tale Trace: Detecting Reasoning Failures in LLMs Using Chain-of-Thought Dynamics
Source: Gary Marcus (X) | 2026-09-02