10,025 views
14/05/2024 Observatory and planetarium Brno www.hvezdarna.cz/program GENERAL ARTIFICIAL INTELLIGENCE (AGI) AND THE DARK SIDE CHATGPT Ing. M.Sc. Jan Romportl, Ph.D. Cybernetic, philosopher, AI consultant The lecture will introduce you to the most important issues of the so-called AI alignment, i.e. efforts to ensure that future general artificial intelligence (AGI) is safe, in line with human values and beneficial to society. The lecture is structured in three parts: 1) Why Large Language Models (LLM) like GPT are definitely much more than just "smart autocomplete" and what are their inherent security problems. 2) A brief history of the big successes of Reinforcement Learning (RL) from 2015-2020, i.e. from AlphaGo to AlphaStar, and why this trajectory stopped for a while. 3) How LLM can bring new fuel to RL, what it means for AGI, why it can be dangerous, and where current AI alignment research is struggling. Visit: https://www.romportl.com/