The AI Breakdown: Daily Artificial Intelligence News and Discussions
OpenAI Sets Moonshot Goal of AI Superalignment in 4 Years
Thu Jul 06 2023Description
The episode discusses changes in data policies and the internet landscape, funding and developments in AI, OpenAI's approach to superintelligence and alignment, as well as reactions and uncertainties about OpenAI's alignment strategy.
Insights
Google's Privacy Policy Update
Google's updated privacy policy allows it to scrape and use data posted online for training AI models, expanding the types of AI that can be used.
Impact of Walled Gardens
The shift from public to private, growth to revenue, and social media to media with comments is changing the internet landscape. Walled gardens give platforms more control over content and can lead to a less open web.
Decline in Funding for AI Companies
Venture capital funding for AI companies has declined by almost half in Q2 compared to last year. This decline may lead to down rounds and lower valuations for previously high-valued companies.
OpenAI's Funding and Developments
AI companies represented 18% of total global funding in the first half of this year, including $10 billion in funding to Open AI led by Microsoft. Entertainment unions like IATSE are adapting to AI and releasing core principles for its applications. India's Tata consultancy plans to upskill engineers on Microsoft Azure's Open in AI.
OpenAI's Approach to Superintelligence
OpenAI acknowledges the possibility of human extinction and aims to build a human-level automated alignment researcher. The moonshot nature of OpenAI's four-year goal is met with mixed reactions from the AI safety community. Concerns are raised about compensation disparities between alignment and capability researchers at OpenAI.
Reactions to OpenAI's Alignment Strategy
Some people appreciate the ambition behind OpenAI's alignment strategy. Dr. Jim Fan suggests moving humans up the supervision chain to better supervise AI systems. Prediction markets show varying levels of confidence in OpenAI's super alignment project. There is uncertainty about what happens if the alignment team doesn't make the desired breakthroughs.
Chapters
- Changes in Data Policies and Internet Landscape
- Funding and Developments in AI
- OpenAI's Approach to Superintelligence and Alignment
- Reactions and Uncertainties about OpenAI's Alignment Strategy
Changes in Data Policies and Internet Landscape
00:01 - 06:19
- Google updated its privacy policy, allowing it to scrape and use data posted online for training AI models.
- This change expands the types of AI that Google can use the data for.
- Twitter and Reddit have also made changes to their API policies to prevent scraping of their data.
- The shift from public to private, growth to revenue, and social media to media with comments is changing the internet landscape.
- Walled gardens give platforms more control over content and can lead to a less open web.
- AI-driven storm is unraveling the public web as sites struggle with AI-generated input and try to protect their data.
Funding and Developments in AI
06:04 - 12:38
- Venture capital funding for AI companies has declined by almost half in Q2 compared to last year.
- The decline in funding may lead to down rounds and lower valuations for companies that raised at high valuations previously.
- AI companies represented 18% of total global funding in the first half of this year, including $10 billion in funding to Open AI led by Microsoft.
- Entertainment unions like IATSE are adapting to AI and releasing core principles for its applications.
- IATSE aims to prepare its members for the future impact of AI through research, collaboration, education, advocacy, organizing, and collective bargaining.
- India's Tata consultancy plans to upskill 25,000 engineers on Microsoft Azure's Open in AI.
- The US military is testing large language models trained on proprietary secret military data to quickly access information and generate new options.
- China plans to curb exports of AI chip-making materials starting from August 1st.
- OpenAI has started a new team called Super Alignment dedicated to steering and controlling superintelligent AI systems.
OpenAI's Approach to Superintelligence and Alignment
12:23 - 18:42
- OpenAI acknowledges the possibility of human extinction and the timeline for superintelligence arrival
- OpenAI aims to build a human-level automated alignment researcher and dedicate a fifth of their compute resources to this goal
- The moonshot nature of OpenAI's four-year goal is met with mixed reactions from the AI safety community
- Concerns are raised about compensation disparities between alignment and capability researchers at OpenAI
- OpenAI's commitment to benchmarks and transparency is seen as a positive step by some, while others remain skeptical
- The ambition behind OpenAI's initiative is recognized and appreciated by many
Reactions and Uncertainties about OpenAI's Alignment Strategy
18:15 - 20:57
- Some people appreciate the ambition behind OpenAI's alignment strategy.
- Dr. Jim Fan suggests moving humans up the supervision chain to better supervise AI systems.
- Prediction markets show varying levels of confidence in OpenAI's super alignment project.
- There is uncertainty about what happens if the alignment team doesn't make the desired breakthroughs.
- OpenAI's effort may influence other companies to follow suit on alignment issues.
- The host invites listeners to share their thoughts and join the conversation on Twitter.