Risks and Benefits of Open Sourcing Language Models
This is a hybrid meeting. The Teams link is https://shorturl.at/pENR0. RSVP (for catering): https://forms.gle/vW1supLXT354JkkU8
This debate focuses on evaluating the risks and benefits of open-sourcing Large Language Models (LLMs). The proponents will argue in favor of open sourcing, aiming to demonstrate that its benefits, including fostering innovation and broad access, exceed its risks. Conversely, the opponents will advocate for keeping LLM research and development under closed source, arguing that the potential negative impacts, such as misuse and security issues, are more significant than the advantages.

The Debaters:
Proposition team:
Christian​ Schroeder
Francisco​ Girbal
Fazl​ Barez
Opposition team:
Jakob​ Foerster
Joshua​ Loo
Aleks​ Petrov
Voting:
The audience will be asked to vote before and after the debate on the key topics using pseudonyms (to track vote-migration).

A few arguments on the two sides:

Topics Easy to Defend from Proposition Side (Pro Open Sourcing)

Accessibility and Democratization: Promoting accessibility and democratization of technology for wider benefit.
Innovation and Collaboration: Accelerating innovation through collaborative efforts.
Transparency and Ethical Development: Ensuring transparency and ethical development of models.
Educational Value and Skill Development: Providing educational value and skill development opportunities.
Global Participation and Diverse Perspectives: Encouraging global participation and diverse perspectives.
Backdoors in Closed-source Models: Arguing that closed-source models can have undisclosed backdoors, making open source models a safer option.
Transparency in Regulation and Public Trust: Advocating that open sourcing enhances regulatory transparency and public trust.
Economic Inclusivity: arguing that open sourcing reduces economic inequalities by providing free access, fostering community support, encouraging entrepreneurship, utilizing global talent, reducing costs for public sectors, spurring economic growth, and building local tech ecosystems.

Topics Easy to Defend for the Opposition (Against Open Sourcing)

Quality Control and Standardization: Ensuring consistent quality and standardization.
Intellectual Property and Commercial Interests: Protecting commercial interests and intellectual property.
Security Concerns: Highlighting risks like misuse by malicious actors or exposure of sensitive data.
Resource Intensity and Sustainability: Addressing the resource demands and sustainability of open source projects.
Technical Challenges and Maintenance: Discussing the technical and maintenance challenges.
Detection of Bad Actors: Emphasizing the difficulty in detecting bad actors in open source projects.
Safety Risks from Highly Capable Models: Highlighting the safety risks associated with open sourcing highly capable models.
Economic Impact: Arguing that open sourcing could exacerbate economic inequalities by benefiting only those with the resources to utilize these models.
Date: 8 February 2024, 17:00
Venue: Thom Building, Parks Road OX1 3PJ
Venue Details: Lecture Room 2
Speakers: Jakob Foerster (University of Oxford), Christian Schroeder (University of Oxford), Aleksandar Petrov (University of Oxford), Francisco Girbal (University of Oxford), Fazl Barez (University of Oxford), Joshua Loo (University of Oxford)
Organising department: Wolfson College
Organisers: Mr Csaba Botos (University of Oxford), Dr. Yi Yin (Wolfson College, University of Oxford)
Organiser contact email address: yi.yin@wrh.ox.ac.uk
Part of: Oxford Cross-Disciplinary Machine Learning (OxfordXML) Research Cluster Seminar Series
Booking required?: Not required
Cost: Free (pizza provided)
Audience: Public
Editor: Yi Yin