Mixture of Experts

Is pre-training a thing of the past? In Episode 34 of Mixture of Experts, host Tim Hwang is joined by Abraham Daniels, Vagner Santana and Volkmar Uhlig to debrief this week in AI. First, OpenAI cofounder Ilya Sutskever said that “peak data” was achieved, does this mean there is no longer a need to model pre-training? Next, IBM released Granite 3.1 with a slew of features, we cover them all. Then, there is a new way to steal AI models, how do we protect against model exfiltration. Finally, can NVIDIA Jetson for AI developers really increase hardware accessibility? Tune-in for more! The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. 00:01 — Intro 00:49— Is pre-training over? 10:25 — Granite 3.1 22:23 — AI model stealing 33:38—NVIDIA Jetson

Is o1 Pro worth the cost? In Episode 33 of Mixture of Experts, host Tim Hwang is joined by Marina Danilevsky, Kate Soule and Vyoma Gajjar. First, the experts debrief the 12 Days of OpenAI. Next, we review some of the top papers in NeurIPS, how are the experts keeping up with all these research papers? Then, we are back with another benchmark, can ARC Prize make AGI more tractable? Finally, Meta announced the launch of Llama 3.3 70B with the promise of 405B performance, can we have our cake and eat it too? Find out more on today’s Mixture of Experts! The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity.

What’s the mystery behind the name ChatGPT refuses to discuss? In Episode 32 of Mixture of Experts host Tim Hwang dives into the hottest topics shaping the AI landscape with an all-star panel: Aaron Baughman, Vagner Figueredo de Santana, and Shobhit Varshney. First, they disect the biggest announcements and takeaways from AWS re:Invent 2024, Amazon’s premier AI event. Next, they talk about overcoming architectural vulnerabilities in AI systems, and finally, they uncover the curious case of a name ChatGPT won’t discuss—and the questions this raises about privacy and transparency in AI. Get ready for an episode packed with insights, debates, and forward-thinking perspectives! The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity.

How much future learning will be done with an AI assistant? In Episode 31 of Mixture of Experts, host Tim Hwang is joined by Phaedra Boinodiris, Marina Danilevsky and Skyler Speakman for the AI in education special episode. First, the experts give an update on the state of AI within education. Next, we cover concerns around AI safety and literacy, what do students and teachers need to be aware of? Finally, the panel gives their predictions on what the future of education holds as it relates to AI. Tune-in to this special episode for an in-depth analysis! The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity.

Should your AI assistant remember everything about you? In Episode 30 of Mixture of Experts, host Tim Hwang is joined by Vagner Santana, Vyoma Gajjar and Shobhit Varshney. First, the experts breakdown claims of “near-infinite memory” within AI models. Next, Shobhit is fresh off the plane from Microsoft Ignite, he shares some of the exciting new announcements following the event. Then, a new benchmark has entered the chat, what do we know about FrontierMath? Finally, AlphaFold3 is now more open, why does this matter? Find out more on today’s episode! The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity.