method · first seen 21 Oct 2026
mutual information maximization
Training objective that increases the statistical dependence between two variables, such as reasoning and output.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Monitorability Is Trainable: Post-Training for Chain-of-Thought MonitorabilityPapers · 21 Oct 2026 · Usman Anwar, Tim Bakker, Christos Louizos, David Krueger
Latest sources
- Monitorability Is Trainable: Post-Training for Chain-of-Thought MonitorabilityPapers · 21 Oct 2026 · Usman Anwar, Tim Bakker, Christos Louizos, David Krueger