Company Updates

OpenAI's AI tools now cost more than half of researchers' salaries, internal data shows

Share
OpenAI's AI tools now cost more than half of researchers' salaries, internal data shows

OpenAI's internal AI tools are now costing the company more than half of its researchers' base salaries, with median daily usage reaching $2,400 in October. The surge in usage highlights the growing dependence on AI agents for research and development, despite their limitations.

TL;DR

  • OpenAI's internal AI tools cost more than half of researchers' base salaries, with median daily usage at $2,400 in October.
  • AI agents are crucial for research acceleration but struggle with complex tasks, requiring human oversight.
  • OpenAI aims to document recursive self-improvement (RSI) as AI systems speed up the development of future generations.

What happened

OpenAI has published internal data showing a significant increase in the use of AI tools like Codex and Dots. In October, the median researcher consumed $2,400 worth of tokens per day, a 62% increase from the previous month. The top 10% of users had a theoretical value of $23,000 per day, double the previous month's figure.

The company's internal tools are running the equivalent of 5.7 workdays per researcher per calendar day, with more than two-thirds of tracked researchers running at least four agents in parallel at peak usage. In September, the theoretical API equivalent of that agent runtime was twice as high as researchers' regular salary costs, excluding equity packages.

The success rate of autonomous agents on routine tasks has improved, with a 46% success rate on two- to four-hour tasks and a 35% success rate on four- to eight-hour tasks. However, agents struggle with more complex engineering work, with success rates as low as 25% for writing code for inference and serving systems.

Why it matters

The growing dependence on AI tools highlights the potential for recursive self-improvement (RSI), where AI systems speed up the development of future generations. This could significantly impact the pace of AI research and development, making it crucial for startups and investors to monitor these trends.

Despite the benefits, AI agents still produce a lot of 'AI slop' and require close human oversight to avoid wasted compute cycles. This underscores the importance of human-AI collaboration in AI research and development.

The rapid advancement in AI tools and their integration into research processes could lead to abrupt advances in deep-learning research itself. This could have significant implications for the competitive landscape and the future of AI development.

Key facts

  • Median daily AI tool usage cost per researcher: $2,400 in October, up 62% from the previous month.
  • Top 10% of users' daily AI tool usage cost: $23,000, double the previous month's figure.
  • AI agents run the equivalent of 5.7 workdays per researcher per calendar day.
  • Success rate of autonomous agents on two- to four-hour tasks: 46%.
  • Success rate of autonomous agents on four- to eight-hour tasks: 35%.
  • Success rate for building evaluation and reinforcement-learning environments: 29%.
  • Success rate for writing code for inference and serving systems: 25%.
  • Internal agent use accounted for 2% of OpenAI's total research compute in September.

Context

OpenAI's internal use of AI tools is part of a broader trend in the AI industry, where companies are increasingly relying on AI to accelerate research and development. This trend is driven by the potential for recursive self-improvement (RSI), where AI systems speed up the development of future generations.

The rapid advancement in AI tools and their integration into research processes could lead to significant changes in the competitive landscape. Companies that can effectively leverage AI tools for research and development may gain a significant advantage over their competitors.

However, the reliance on AI tools also raises concerns about the potential for job displacement and the need for close human oversight. As AI tools become more advanced, it will be crucial for companies to strike a balance between automation and human input in their research and development processes.

Topics

Join the discussion

Have a take on this story? Weigh in with our community on Facebook.

💬 Discuss on Facebook →