According to the source, a new study proposes a simple pre-prompt that coaxes alignment-trained large language models to reveal multiple possible answers instead of a single response, restoring diversity reduced by alignment training (source: X post dated Oct 15, 2025). According to the source post, no model list, benchmark metrics, peer-review status, or market impact data were provided, so there is no direct evidence yet of price effects in AI-linked crypto or equities (source: X post dated Oct 15, 2025).