EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
MIT Technology Review·Aug 18·5 min read
We still don’t know how people are really using AI
TL;DR
The AI Observatory, a new research initiative, aims to provide independent insights into how people use AI models like Claude and ChatGPT, revealing significant gaps in existing data. Its findings indicate that AI usage varies widely and includes more sensitive behaviors than reported by major companies.
✦ Why It Matters
Consider using the AI Observatory's findings to inform your AI-related decisions and policies today.
Key Takeaways
- Explore the AI Observatory for comprehensive AI usage data.
- Recognize the limitations of company-released AI usage reports.
- Investigate how different AI models cater to various user needs.
- Stay informed about sensitive topics in AI interactions.
Full Summary
AI companies often release selective reports on how their products are used, leaving a gap in understanding actual user interactions. The AI Observatory, co-led by Anka Reuel from Stanford, aims to address this by aggregating 24,521 conversations from seven datasets, providing a broader view of AI usage.
The research found that nearly half of conversations would be filtered out by existing company metrics, revealing more sensitive topics like health and relationships. Additionally, the Observatory noted that AI use varies significantly across models, with Grok favored for news and politics, while ChatGPT is used for homework assistance.
The findings suggest that current company reports do not capture the full spectrum of AI interactions, emphasizing the need for independent research to inform policy and development.