
Niamh ('Neev') Gavin
Applied AI Scientist
Talk
AI Research Fails - Anecdotes from the Frontline
GPUs went Brrrr then... This session will illustrate the 'law of unintended consequences' in AI research, using lesser known examples from pre/post/prod that caused the model to 'go rogue' in unexpected ways. We'll then run through the tooling used to root cause, (re)align and monitor at scale.
About
Niamh is an Applied AI Scientist. With OpenAI and DeepMind she optimised research for commercial applications, and for open models she enabled low latency releases of Llama (3.2 70b) and Mistral (7b, 8x16). Niamh's academic background is AI for scientific discovery; NatSci, SymSys, CompBio (Oxford, Stanford, Tsinghua).