Papers

Download PDF (preprint v1.2, 10 pages) LaTeX source Superseded: v1.1 PDF, v1.1 source, DOI 10.5281/zenodo.22882691 DOI: 10.5281/zenodo.22960030 (this version, v1.2) · 10.5281/zenodo.22882690 (all versions) · License CC BY 4.0 Version 1.2 (2026-09-25) corrects version 1.1 (2026-09-21). Two of the 34 runs v1.1 ranked, claude-opus-4.7 and claude-sonnet-4.6 through the relay api.b.ai, had their replies re-cased to upper case by the relay, which the canon validators reject; v1.1 took the near-zero scores for the models’ own failure. The same requests through a second relay came back in lower case with the same content. Those runs are excluded as a fifth class of relay interference, the comparable set is 32 runs, and every dependent figure is recomputed (median grammar 0.83, median execution 0.11, R9 share 93.4%, boundary accuracy 55%). See the corrected results dataset. ...

Abstract Current approaches to improving LLM performance focus on expanding context windows and augmenting memory. This paper argues the opposite: strategic forgetting — selectively pruning low-value information from context — produces better outcomes than unbounded memory accumulation. We propose a formal framework for selective forgetting and demonstrate its effectiveness in multi-session AI interactions. Status In preparation.

Download PDF (v0.5) Abstract AI hallucination is not an engineering bug — it is the epistemological ceiling of inductive reasoning applied to non-stationary environments. This paper traces the problem from Hume’s Problem of Induction through PAC learning bounds to the No Free Lunch theorem, proving that zero hallucination is mathematically impossible in open systems. The root cause is not insufficient data or compute, but the fundamental limitation of pattern-matching on historical observations to predict novel futures. ...