Workshop on AI Scaling and its Implications

The FutureTech workshop on AI scaling and its Implications aims to gather computer scientists, engineers, economists, and policy researchers to think about scaling laws and their implications, such as:

How far can scale take us?

What do scaling laws predict about requirements for mastery of a range of key tasks? Will the scaling of current systems (with conservative extensions) be sufficient?

Investigating future AI systems

How large will such models be? Will future systems be highly general or more specialized than current systems?

Scaling and Automation

What are the predictable impacts of automation and the prospects for accelerated economic growth?

The Scaling Runway and Beyond Neural Scaling

How long can current scaling trends be sustained? Can this "runway" be extended? Are there more efficient paths to advanced AI?

Speakers

Jacob Steinhardt

UC Berkeley

Pamela Fine Mishkin

OpenAI

Phillip Isola

MIT

Rylan Schaeffer

Stanford

Eric Drexler

University of Oxford

Anton Korinek

Brookings, Centre for the Governance of AI

Joel S. Emer

MIT

Dan Hendrycks

Center for AI Safety

Philip Trammell

University of Oxford

Adam H. Marblestone

Convergent Research

Max Tegmark

MIT

Jonathan S. Rosenfeld

MIT

Carl Shulman

Open Philanthropy

Owain Evans

University of Oxford

Hailey Schoelkopf

EleutherAI

Agenda

Thursday, October 12th

8:15 AM

Breakfast

9:00 AM

Kickoff

9:15 AM

The state of scaling laws today

Jonathan S. Rosenfeld

MIT

10:00 AM

Scaling and downstream capabilities

Rylan Schaeffer

Stanford

11:00 AM

Break

11:30 AM

Scaling and generalization failures

Owain Evans

University of Oxford

12:00 PM

Scaling, Automation, and Growth

Philip Trammell

University of Oxford

12:45 PM

Lunch

1:45 PM

AI scaling and interpretability

Max Tegmark

MIT

2:30 PM

AI automation and wages

Anton Korinek

Brookings, Centre for the Governance of AI

3:15 PM

Break

3:35 PM

How we got here and what's next in scaling, parallelism, and hardware

Hailey Schoelkopf

EleutherAI

4:10 PM

Role Architectures: Scalable Competence Is All You Need

Eric Drexler

University of Oxford

6:30 PM

Dinner at Catalyst

Friday, October 13th

8:15 AM

Breakfast

9:00 AM

Kickoff

9:15 AM

Hardware innovations driving AI scaling

Joel S. Emer

MIT

Scaling generative models and synthetic data

Phillip Isola

MIT

10:00 AM

10:45 AM

Break

11:30 AM

Automation and large language models

Pamela Fine Mishkin

OpenAI

12:15 PM

Scientific progress and AI automation

Adam H. Marblestone

Convergent Research

1:00 PM

Lunch

2:15 PM

Research priorities discussion

2:30 PM

AI, Natural Selection, and Competition

Dan Hendrycks

Center for AI Safety

3:15 PM

The productivity of AI workers

Carl Shulman

Open Philanthropy

4:00 PM

What will GPT-2030 Look Like?

Jacob Steinhardt

UC Berkeley

Locations

MIT Museum

Gambrill Center, 314 Main St, Cambridge, MA

Legal Sea Foods - Cambridge

355 Main Street, Cambridge, MA

Catalyst Restaurant

300 Technology Square, Cambridge, MA

Contact

This workshop is organized by the FutureTech team (Neil Thompson, Tamay Besiroglu, Carl Guo, and Shiobhain Warren-Jenkins).
For questions or more information, reach out to scaling-workshop@mit.edu.