I’ve argued for “hallucination” → “Runge Spikes”; see https://news.ycombinator.com/item?id=43612517
“Runge Spikes” is still a metaphor, rather than a technically accurate description. Its strength is that it gets away from using a metaphor from human illness. Metaphors from human illness smuggle in an assumption that LLMs are faulty humans, rather than technology doing its thing.
There is logical gap. Earlier in the piece:
When a conflict theorist encounters a statement, they ask: “What is this statement meant to achieve in the world?”
But later:
Conflict Theory/ Post Modernism: Believe what is useful but ask first useful to whom?
The piece starts by looking at intentions. The statement is meant to achieve X. The speaker has a theory of how the world works and shapes what they say to persuade others to behave in ways that theory says will lead to X. But their theory is wrong; they succeed in persuading others and achieve Y instead. What the statement is meant to achieve is X, and we are focusing on X and neglecting Y.
Later though, we look at usefulness, and must resolve the ambiguity: usefulness of X or usefulness of Y?
This ambiguity bleeds back into the concept of mistake theory. Consider the plight of a mistake theorist whose preference order over outcomes is X is better that Z is better than Y, and who also possesses a theory, both disappointing and true, that warns that X is not a possible outcome. The mistake theorist argues for Z, hoping to be spared Y. The conflict theorist argues for X, and dismisses warnings from the mistake theorist. The conflict theorist bases his dismissal on noticing that warning “X is impossible” is useful to those who prefer Z to X. The conflict theorist may be correctly understanding a different mistake theorist who does indeed prefer Z to X, and jumps happily on the “X is impossible” band wagon. The conflict theorist hasn’t understood the motivation of the first mistake theorist, who prefers Z to Y.
For a concrete example, consider the Laffer curve. Are we on the part of the curve where increasing tax rates brings in more revenue? Or will increasing tax rates bring in less revenue? I’ll try to shorten this comment by only considering the latter case.
There is a three way fight. (A) Those who favour lower tax rates, despite an erroneous belief that higher tax rates will bring in more revenue. (B) Those who favour higher tax revenues and reject higher tax rates based on an accurate assessment of the Laffer curve. (C) Those who favour higher tax revenues and mistakenly advocate for higher tax rates to gather those revenues.
The C’s correctly discern both that the A’s are arguing that society is over the peak of the Laffer curve, and that the A’s do not in fact believe it themselves; they merely want lower tax rates. The C’s win power. Higher tax rates. The future arrives. Falling tax revenues, bond market crisis, IMF, cuts to the spending that the C’s see as the essence of civilised society.
Now we get to revisit the issue of usefulness. It turns out that Laffer curve argument would have been useful to C’s but they rejected it partly because it was useful to their enemies and partly because the C’s were mistaken.
Maybe some mistake theorists are seeking truth across the board, but other mistakes theorists just want to avoid becoming collateral damage in conflict-theorists’ unforced errors.