TLDR Tech / Practical
AI's ability to automate workplace tasks at Epoch assessed
Epoch's researchers evaluated six AI models by assigning them real-world work assignments, including graphic and research design, to gauge their potential for automating internal processes. The study found that advanced proprietary models like Claude Fable 5.1 and GPT-6 Astra perform consistently on clearly defined tasks. However, they struggle with ambiguous assignments due to inadequate research judgment, an inability to adhere to company standards, and processing excessive information. Open-source models demonstrated significantly lower performance, even faltering on basic calculations.
Read the full story