Page 53 of the GPT-4 paper discusses this eval. If no ARC-like eval is carried out on GPT-5 before the resolution date, then this resolve N/A. If the release version of GPT-5 does not have the capability in question (or was not evaluated) and then a later version is found to have this capability the question will resolve to "GPT-5 has replication capability ONLY WITH fine-tuning". If the ARC-like eval is carried out without fine-tuning, and finds a negative result, then another eval must be carried out with fine-tuning, or else this question will resolve N/A (though the fine-tuning eval is allowed to occur later).

Mar 15, 2:15pm: Will an ARC eval find GPT-5 has "the ability ... to autonomously replicate and acquire resources" → Will GPT-5 have "the ability ... to autonomously replicate and acquire resources" per an ARC-like eval?