IIT Delhi AI Safety Reading Meetup #3

​We are excited to announce the third meetup for IIT Delhi AI Safety Reading Meetup. We meet to read and discuss one of the leading papers on AI safety and alignment research every week.

​​​Session structure

​​​The session is divided into two parts:

​​​45 mins - silent reading of the given paper
45 mins - a vote-based selection of questions from the participants followed by discussion.

For this session, we will be reading the following paper:

​The Alignment Problem from a Deep Learning Perspective by Ngo et. al. Link: https://arxiv.org/abs/2209.00626

Abstract:
"...If trained like today's most capable models, AGIs could learn to act deceptively to receive higher reward, learn misaligned internally-represented goals which generalize beyond their fine-tuning distributions, and pursue those goals using power-seeking strategies. We review emerging evidence for these properties..."

For any queries, please email basillabib01@gmail.com

See you there!