LLM safety benchmarks evaluate one response at a time.
New research from Vector Faculty Affiliate @edolatabadi and Faculty Affiliate Researcher Abeer Badawi asks what happens across the full arc of a conversation and whether AI mental health tools quietly erode human agency.
🧵
Vector Institute transforms cutting-edge artificial intelligence research into practical solutions. AI-generated content will be disclosed. FR: @InstitutVecteur
- Vector's 2025-26 Annual Report just dropped! Our research community earned significant recognition this year, with Faculty Members @A_Aspuru_Guzik, @shaibendavid5, and @BoWang87 elected Fellows of the Royal Society of Canada and Vector Faculty member @NicolasPapernot receiving
- RBC announces renewed sponsorship with Vector 🎉 Nine years ago, @RBC was among the first organizations to join Vector to advance AI research, innovation, and talent development across the country. We’re excited to continue our collaboration and advance AI for Canada. Read the
- Somewhere around layer 28 of a 35-billion-parameter multimodal model, something changes. In the early layers, the router allocates compute based on signal type, audio, video, and text. In the deeper layers, it routes by meaning. The same input takes a different path depending on
- We’re excited to share that Vector is receiving a gift of Claude API credits from @AnthropicAI as part of a wider $10 million CAD commitment to Canadian AI research. Through this donation, Vector researchers will have access to these credits to further advance their work in AI

