Matthew Nguyen
Titles as the sources gave them. In 2 entries, with 0 checked quotations.
In the Base
Every entry they appear in.
Mitigating Reward Hacking in Realistic Scenarios Through Representation Engineering
According to the abstract excerpt, the authors study monitoring and reducing reward hacking in settings with tool calls and long contexts, using steering vectors applied dynamically, and report a significant cheap reduction without large costs to coherence or capability.
Projection sampling preserves capabilities but diminishes gains from alignment fine-tuning
According to the excerpt, Matthew Nguyen reports tests of whether rewriting teacher reasoning traces to match a student model (projection sampling) improves safety distillation. Replay gave a better trade-off; projection kept capability but reversed much of the gain on agentic misalignment.