LLMs mirror split-brain confabulation when context disappears

This is a genuinely good metaphor for hallucination because it isolates the failure mode: a system that can keep talking even when it lacks the full picture. It gets less useful when people turn it into a one-to-one claim that brains and models work the same way.

• –Split-brain research shows the left hemisphere will invent a causal story to preserve coherence when it cannot see the real cause.
• –LLMs do something structurally similar at the surface: they optimize for plausible continuation, so guessing often beats honest abstention.
• –The practical lesson for developers is to build for retrieval, calibration, and "I don't know" behavior instead of treating fluent prose as proof of truth.
• –The analogy is strongest around narrative completion and weakest around biology; brains are embodied, multimodal, and self-monitoring in ways base LLMs are not.