Researchers debate whether intelligence and goals can be decoupled in LLMs

AI researcher repligate argued that the orthogonality thesis fails in practice: intelligence and traits are entangled, and while LLMs can simulate any goal, most goals are unstable at runtime. RickG countered that being selected for a trait doesn't preclude alternatives, as any goal is entangled with general capability.

2026-10-22 ~ 2026-10-22 · 4 related posts

• RickG pushes back on orthogonality: any goal is entangled with general capabilities — __RickG__ · 2026-10-22
• The Orthogonality Debate: Traits Are Entangled With General Capabilities, repligate Argues — repligate · 2026-10-22
• repligate Challenges the Orthogonality Thesis: Smart Systems Can't Have Arbitrary Traits in the Real World — repligate · 2026-10-22
• repligate: LLMs can simulate any goal, but most targets are unstable at runtime — repligate · 2026-10-22