Rich Sutton warns against relying on synthetic data for AI training

Key Takeaways

Rich Sutton Warns Against Relying on Synthetic Data for AI Training

AI pioneer Rich Sutton says using synthetic data for AI training is 'a big mistake.' He believes it's not the solution for scaling AI. Sutton's comments come as the use of synthetic data grows in the AI industry. Many companies use synthetic data to train AI models when real data is hard to find or expensive to collect. However, Sutton thinks this approach has limitations.

Anterior Discusses Synthetic Healthcare Data for AI

Anuj Iravane of Anterior talks about generating synthetic data to overcome challenges in healthcare AI. The company uses synthetic data to reverse inference workflows and empower clinicians. This approach helps address issues with protected health information (PHI) in healthcare AI. Anterior's method allows for more efficient and effective use of AI in healthcare.

Angana Jacob on AI-Driven Investment Research

Angana Jacob, Head of Research Data at Bloomberg, discusses AI-driven investment research workflows. She highlights the importance of getting data foundations right for AI-driven research. Jacob also talks about 'agentic research workflows,' the latest AI evolution. This approach aims to make data more connected and decision-ready across critical workflows.

NCI Launches AI and Cybersecurity Degree Programs

The National College of Ireland (NCI) is introducing two new degree programs in AI and cybersecurity. The programs aim to meet the growing demand for skills in these areas. Students can expect to learn about AI, digital security, and data-driven technologies. The courses are set to launch in September 2027, subject to approval.

New Hampshire's AI Regulatory Challenges

New Hampshire lawmakers have been slow to regulate artificial intelligence (AI). Several bills were proposed, but none made it to the governor's desk. The state's approach to AI regulation is cautious, with a focus on addressing specific issues like child exploitation. This comes as AI technology continues to evolve and pose new challenges.

OpenAI Revokes Access to Limited Cyber Program

Several security researchers say OpenAI revoked their access to a limited-access program for cybersecurity research. The program, called TAC, provides access to AI models for defensive security work. OpenAI confirmed that the issue was caused by an error and is working to resolve the problem. The incident highlights the challenges of balancing security and access to AI technology.

Capability-Driven Data for Generative AI

Researchers have developed a new capability-driven data infrastructure for generative AI. This infrastructure enables the creation of large-scale multimodal diffusion models. The approach uses specialized data engines and curriculum scheduling to improve the efficiency of AI model training. The methodology has been demonstrated to be effective in training large-scale AI models.

Japan to Require AI Firms to Disclose Training Data

Japan is considering a nonbinding code for generative AI businesses to disclose their training data and methods. The code aims to protect intellectual property rights and promote transparency in AI development. The government will use a 'comply or explain' approach to encourage businesses to follow the code. This move comes as concerns about AI and copyright infringement grow.

Editorial: Tom Tiffany's Ties to AI Oligarchs

An editorial criticizes Tom Tiffany's record on data centers and AI regulation. The editorial argues that Tiffany prioritizes the interests of the data center industry over those of Wisconsin residents. This comes as the state considers the impact of data centers and AI on the environment and local communities.

Europe's AI Data Centers Seek Cheaper Energy and Land

Developers of AI data centers in Europe are seeking locations with cheaper energy and land. The demand for data storage and processing is growing, driven by AI and digital technologies. To meet this demand, developers are looking for less traditional locations with faster and more reliable internet connectivity. The use of renewable energy sources is also becoming more widespread in the data center industry.

Sources

• Using synthetic data for AI training is 'a big mistake': Rich Sutton
• Anterior's Anuj Iravane on Synthetic Healthcare Data
• Data in Focus: Angana Jacob on AI-driven investment research | Insights
• NCI to launch two bachelor’s courses in AI and cybersecurity
• Amid existential threats and giant challenges, NH’s leaders frolic in fields of small potatoes
• Researchers complain that OpenAI revoked their access to limited cyber program
• Capability-Driven Data for Generative AI
• Japan to require AI firms to disclose training data
• Editorial | Tom Tiffany serves the AI oligarchs
• Europe AI data centres seek cheaper, quicker energy and land