External prompt and output filters cannot be both efficient and safe for all language models, even with black-box access, assuming time-lock puzzles exist.
Note that Sol computes s using t repeated squarings, each of which is inexpensive, but the full process requires Θ(t) sequential steps
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
On the Impossibility of Separating Intelligence from Judgment: The Computational Intractability of Filtering for AI Alignment
External prompt and output filters cannot be both efficient and safe for all language models, even with black-box access, assuming time-lock puzzles exist.