AI fixes AI in MIT study, Mistral's open model & Google Genie 4
Show notes
Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts
How close are we to AI that can improve itself?
This week on Mixture of Experts, guest host Aili McConnon is joined by Chris Hay, Sascha Brodsky and Kiyu Gabriel tackle that question and more from the latest AI headlines.
First, MIT researchers demonstrate a technique that uses AI to modify another model's weights. The panel considers how close this brings us to recursive self-improvement and whether reward hacking poses a more immediate risk.
Next, Mistral releases a trillion-parameter open-weight model while NVIDIA expands its open model offerings. With open models approaching frontier-level capabilities, will developers still need to rely on proprietary AI?
Finally, Google DeepMind introduces Genie 4, its new world model. The panel debates how much these models actually understand about physics and what happens when a convincing simulation gets the real world wrong.
00:00 – Intro
01:08 - AI self-improvement and MIT's research
16:28 - New open models from Mistral and NVIDIA
25:27 - Google’s Genie 4 simulates 3D worlds
All that and more on 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. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by the production team before publication."