LLMs and Alignment

Since November 2024, AaltoAI and Tutke have been co-hosting a weekly reading group to track recent developments in large language models through in-depth discussion of research papers.

Over dozens of sessions, we've worked through the foundations of modern AI, like word embeddings, transformers, and scaling laws to open problems in alignment, deceptive behaviour, AI control, and mechanistic interpretability. We've hosted research talks, run collaborative coding sprints, and screened films exploring what intelligence means.

As AI systems become more autonomous and start making decisions that affect the real world, aligning them with human values is more important than ever. Keeping pace with both safety research and capability advances is why we meet each week.

Join us to discuss new research, share ideas, and stay up to date with the rapidly evolving world of AI.