La Revue de Presse — AI safety & alignment

Viktor et Nia discutent des actualités du jour.

Sources :

• LessWrong: Could Frontier AI Researchers Collectively Slow the Race? A Conditional Pledge Mechanism
• LessWrong: The Goblins Are the Paperclips
• Alignment Forum: Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
• LessWrong: Do capabilities generalize across propensities?
• LessWrong: Second order thoughts on current AI agents
• LessWrong: International Law Cannot Prevent Extinction Either