Philip Torr

Subtracting information from the student rather than adding it to the teacher can yield performance improvements that rival those achieved with privileged data.

MMDiff reveals that multimodal SAEs can be powerful tools for both auditing and steering MLLM behavior, achieving up to 24% reduction in safety attack success rates without compromising performance on visual question answering.

Early unification in multimodal training can prevent models from becoming overly reliant on language, unlocking new efficiencies in generative performance.

Simulating 8.3 billion diverse personas reveals nuanced user interactions that traditional evaluations miss, transforming how we assess AI systems.

AI scientists excel at idea generation but falter in filtering and prioritizing innovations, revealing a critical gap in their capabilities.

Structured offline learning can dramatically enhance decision-making in oncology clinical trials, with agents outperforming traditional tools by a significant margin.

None of the 30 LLM agents evaluated in CausalGame demonstrated reliable causal thinking, revealing a critical gap in AI's ability to perform scientific reasoning.

Transformers can be rigorously evaluated for their cryptographic capabilities, revealing upper bounds on their computational power that could redefine security in AI systems.