I enjoyed the Defence AI Symposium 2023, which offered a diverse perspective on using AI in a defence context, emphasizing trustworthiness and responsibility.

I delivered a talk titled “Beyond AI Models: Responsible AI in the Era of Frontier Models,” where I focused on:

1. Exo-model thinking: At the system level, what can we do for assurance, guardrails, and understanding, assuming we will never fully comprehend the model?
2. Process rubrics: How can we move beyond simple yes/no process checks to a deeper evaluation of process quality?
3. Federated LLMs/RAGs: How can we harness the power of frontier AI models developed by others and collaborate with others while maintaining sovereignty in data, intelligence, and intelligence sources?

Selected Slides: Dropbox

𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 at CSIRO’s Data61: https://lnkd.in/gyzjE4-i
𝐁𝐨𝐨𝐤: https://lnkd.in/gsQz5swy