A mathematical ceiling limits generative AI to amateur-level creativity | Hacker News Reader

A mathematical ceiling limits generative AI to amateur-level creativity

psypost.org

16 pointsgeox4 comments

The conclusion of this article is false. There are many counterexamples to it. Just one example - an AI generated song topped the Spotify chart https://www.forbes.com/sites/conormurray/2025/11/20/apparent...

top pop is and has been formulaic and entirely the product of commitees for many decades, with a very significant number of "hits" never actualy performed live , and those that are, are anything but "creative", bieng 100% scripted, and micro stage managed, with lip sinking bieng rampant. I forget, and dont care to remember the widly known limits on tempo, key, and chord change's. That fully boxing and computerising it has happened now, is just an inevitability realised. The last bit is that "poularity" is very much a product, and has essentialy no basis in an organic demand.....you cant like something, you never hear, and what people get to hear is decided, for them. So "spotify" decides to push something to the top of the chart, and looky there, it happens.

By expressing this relationship through a mathematical formula, the study identified a specific upper limit for AI creativity. Cropley modeled creativity as the product of effectiveness and novelty. Because these two factors work against each other in a probabilistic system, the maximum possible creativity score is mathematically capped at 0.25 on a scale of zero to one.

'It's just autocorrect', engineering version

The paper's reasoning appears to be:

1. Creativity = Effectiveness × Novelty, all ranging from 0 to 1

2. In LLM outputs, effectiveness/novelty are negatively correlated (higher-scored continuations have higher effectiveness but lower novelty, and vice versa)

3. "the negative correlation of E[ffectiveness] and N[ovelty] means that Novelty [...] can be expressed as one minus the Effectiveness"

4. So the creatvitiy equation becomes Creativity = Effectiveness × (1 - Effectiveness), which maxes out at 0.25 creativity

5. 0.25 creativity corresponds to the boundary between amateur and professional creativity ("the 4Cs representation of creativity")

But just because there's a negative correlation between E and N does not mean that N = 1 - E. The exact correlation wouldn't matter if these were just arbitrary units with no defined scale or zero-point, but the conclusion rests on referencing "0.25 creativity" against an existing scale to determine its meaning. That there's a negative correlation/trade-off between the two (straying from well-established solutions increases risk of failure) doesn't seem unique to LLMs.

Reply on news.ycombinator.com