The Stanford Institute for Human-Centered Artificial Intelligence (HAI) has unveiled its latest AI Index Report, providing an in-depth analysis of the artificial intelligence industry's current state. The 502-page document, compiled by experts from both academia and industry, reveals a sector that is flourishing yet grappling with escalating costs, increasing regulations, and rising public apprehension. The report emphasizes the importance of transparency in how large language models (LLMs) use data, acknowledging the challenges in obtaining genuine and informed consent for data collection.
Costs and Open Source Models
One of the report's significant findings is the sharp increase in the costs associated with developing frontier AI models. The median training costs have nearly doubled in the past year, with state-of-the-art models reaching unprecedented expenditure levels. For instance, OpenAI's GPT-4 required an estimated $78 million worth of compute resources, while Google's Gemini Ultra demanded $191 million. Despite these soaring costs, the report notes a trend toward more open-source models, with 65.7 percent of the 149 foundation models released in 2023 being open source, up from 44.4 percent in 2022.
Regulations and Future Outlook
The AI Index Report also highlights the sharp increase in AI-related regulations in the United States, rising from one in 2016 to 25 in the last year. These regulations are expected to introduce additional costs and compliance challenges for AI firms. Moreover, the report points to a growing public concern over AI's potential impact on society, with an increase in the number of people feeling apprehensive about AI technologies. Looking forward, the report discusses the dual futures of AI: one where technology continues to advance and integrate into various sectors, and another where its adoption is limited by technological constraints and societal concerns.
