[jacobgkau]

The question is whether humans specifying the relationships between nodes is more or less reliable than LLMs inferring those relationships.

[svachalek]

So we need to compare the reliability of something hypothetical to something real? I'll take the real every time.

[jacobgkau]

What? No. With traditional tech stacks, humans are responsible for tagging and structuring data as they input it. With AI, the argument is that the AI system can do that instead (via inference), meaning humans can input it unstructured. They're both "real."

If you were trying to make a witty comment about AI's purported capabilities not existing/working well, then I personally agree that humans doing it would be smarter and more efficient.