Archive — US — techcrunch
OpenAI says GPT-5 stacks up to humans in a wide range of jobs
OpenAI released GDPval, a benchmark testing GPT-5 and Anthropic's Claude Opus 4.1 against human professionals. While GPT-5 performed comparably in 40.6% of tasks, improvements indicate AI's increasing capabilities. However, GDPval currently only covers limited tasks. OpenAI sees potential for AI to assist professionals in high-value tasks but acknowledges the need for more comprehensive testing to measure real-world applications effectively.
Key points
- OpenAI introduces GDPval benchmark to compare AI performance (GPT-5, Claude Opus 4.1) with human experts across nine key US industries.
- GPT-5 demonstrated a win rate of 40.6% against industry experts, while Claude Opus 4.1 achieved 49%.
- GDPval tests primarily involve report generation, which is a narrow representation of professional tasks.
- OpenAI acknowledges that current tests do not encompass the full scope of human jobs.
- Plans for more extensive testing in the future to include a wider range of industries and tasks are in place.
Why it matters
- This news relates to ongoing discussions about AI's role in the workforce, especially predictions of AI replacing certain job categories.
- The launch of GDPval reflects the industry's need for robust benchmarks as AI technologies adopt more complex real-world functions.
- As we approach 2025, trends show a push for AI models to integrate into various fields seamlessly, impacting job roles and skills required.
OpenAI's GDPval reveals promising advancements in AI, though limited task scope highlights the complexity of real-world applications; further, broader assessments are necessary for a complete evaluation of AI's potential in the workplace.