Source
Futurism
Summary
OpenAI published a new evaluation, GDPval, assessing how well its models perform “economically valuable” tasks across 44 occupations. The results suggest that current frontier models are approaching the quality of expert work in many domains. Examples include legal briefs, marketing analyses, technical documentation, medical image assessments, and sales brochures. While AI might not replace entire jobs, it can outperform humans in well-specified tasks. OpenAI emphasises that models currently handle repetitive, clearly defined tasks better than nuanced judgment work. GPT-5-High matched or surpassed expert deliverables in ~40% of evaluated cases. Critics warn of hallucinations, overconfidence, and the risk of overestimating AI’s real-world reach.
Key Points
- GDPval tests 44 occupations on real-world tasks to benchmark AI against experts.
- GPT-5-High achieved parity or better than expert work in ~40% of tasks.
- Tasks include analytics, document drafting, medical imaging, and sales collateral.
- AI models perform best on repetitive, narrow tasks; struggle on ambiguous, poorly defined ones.
- OpenAI positions this not as job replacement but augmentation—yet raises deeper questions about labour, oversight, and trust.
Keywords
URL
https://futurism.com/future-society/openai-work-tasks-chatgpt-can-already-replace
Summary generated by ChatGPT 5
