The widepread deployment of AI systems in critical domains demands more rigorous approaches to evaluating their capabilities and safety.
The widepread deployment of AI systems in critical domains demands more rigorous approaches to evaluating their capabilities and safety.
2025 Spring Conference
In an era when information is treated as a form of power and self-knowledge an unqualified good, the value of what remains unknown is often overlooked.
In an era when information is treated as a form of power and self-knowledge an unqualified good, the value of what remains unknown is often overlooked.
HAI Weekly Seminar
The question of whether machines can really understand. Lovelace and Searle thought not. Turing thought yes. (Ok that’s not totally accurate, but let’s not ruin a good story.) With the advent of LLMs the question has resurfaced in force, again with some strong skeptics such as Bender et al. In this seminar, Yoav Shoham shares his views, based mostly on work at AI21 Labs. Spoilers: (1) He’s with Turing. (2) LLMs, as currently built, are necessary but not sufficient. (3) The question is more interesting than the answer.
Professor Emeritus of Computer Science, Stanford University
No tweets available.