Posted on 4th September 2024 by Harry
Zico Kolter is a Professor and the Director of the Machine Studying Division at Carnegie Mellon College. His analysis spans a number of matters in AI and machine studying, together with work in AI security and robustness, LLM safety, the influence of knowledge on fashions, implicit fashions, and extra. He additionally serves on the Board of OpenAI, as a Chief Knowledgeable for Bosch, and as Chief Technical Advisor to Grey Swan, a startup within the AI security house.
In Right now’s Episode with Zico Kolter We Focus on:
1. Mannequin Efficiency: What are the Bottlenecks:
- Knowledge: To what extent have we leveraged all out there knowledge? How can we get extra worth from the info that we’ve got to enhance mannequin efficiency?
- Compute: Have we reached a stage of diminishing returns the place extra knowledge doesn’t result in an elevated stage of efficiency?
- Algorithms: What are the most important issues with present algorithms? How will they modify within the subsequent 12 months to enhance mannequin efficiency?
2. Sam Altman, Sequoia and Frontier Fashions on Knowledge Centres:
- Sam Altman: Does Zico agree with Sam Altman’s assertion that “compute would be the forex of the long run?” The place is he proper? The place is he improper?
- David Cahn @ Sequoia: Does Zico agree with David’s assertion; “we are going to by no means prepare a frontier mannequin on the identical knowledge centre twice?”
3. AI Security: What Folks Assume They Know However Do Not:
- What are individuals not involved about in the present day which is an enormous concern with AI?
- What are individuals involved about which isn’t a real concern for the long run?
- Does Zico share Arvind Narayanan’s concern, “the most important hazard isn’t that individuals will consider what they see, it’s that they won’t consider what they see”?
- Why does Zico consider the analogy of AI to nuclear weapons is improper and inaccurate?
