Brian - I had no idea Anthropic had data on social workers. Fascinating. I'm not sure I understand some of the data. For example, what does it mean that 100% participate in student recruitment, registration, and placement activities? Does that mean that of the conversations that social workers have using an API 100% that addressed administrative tasks were completed by AI without human editing? You wrote, "This selective approach of maximizing automation for administrative tasks while preserving human judgment for educational functions distinguishes social work education from sectors pursuing uniform automation strategies." I want to understand, but I'm confused.
Hey Jonathan: The Anthropic report is limited with its reporting on methodology, so I cannot say for certain. My interpretation of the data is consistent with yours. That is, conversations with the AI with respect to human recruitment, registration, and placement are directive. According to Anthropic's definition of being directive, that means there is almost no human in the loop.
I found it surprising that there were multiple social work categories in the report. I'm also not entirely sure how they applied the social work label to conversations. Presumably, they are using a classification model on user conversations, as this level of detail is not collected when people register for their service.
I haven't seen any report by the major LLM companies, so this one caught my eye when it was distributed. With respect to economic impacts of AI on the labor market, the work of Felten (Princeton University) is very important. He focuses on the concept of LLM "exposure". This is defined as the extent to which an occupation involves tasks that AI systems can potentially perform, based on matching AI progress metrics with occupational ability requirements from O*NET data. His earlier reports following the initial release of ChatGPT placed social work educators in the top 20 jobs with the greatest exposure. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4375268
Brian - I had no idea Anthropic had data on social workers. Fascinating. I'm not sure I understand some of the data. For example, what does it mean that 100% participate in student recruitment, registration, and placement activities? Does that mean that of the conversations that social workers have using an API 100% that addressed administrative tasks were completed by AI without human editing? You wrote, "This selective approach of maximizing automation for administrative tasks while preserving human judgment for educational functions distinguishes social work education from sectors pursuing uniform automation strategies." I want to understand, but I'm confused.
Hey Jonathan: The Anthropic report is limited with its reporting on methodology, so I cannot say for certain. My interpretation of the data is consistent with yours. That is, conversations with the AI with respect to human recruitment, registration, and placement are directive. According to Anthropic's definition of being directive, that means there is almost no human in the loop.
I found it surprising that there were multiple social work categories in the report. I'm also not entirely sure how they applied the social work label to conversations. Presumably, they are using a classification model on user conversations, as this level of detail is not collected when people register for their service.
Ok. Thank you. Good to know. Do you know if any of the other LLM companies (e.g. OpenAI, Google, Microsoft) publish a similar report?
Outside of the Economic Report, there are environmental impact reports by Mistral and Google. I'm very impressed by the Google report, as it quantifies the energy use of a single median query. https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference
I haven't seen any report by the major LLM companies, so this one caught my eye when it was distributed. With respect to economic impacts of AI on the labor market, the work of Felten (Princeton University) is very important. He focuses on the concept of LLM "exposure". This is defined as the extent to which an occupation involves tasks that AI systems can potentially perform, based on matching AI progress metrics with occupational ability requirements from O*NET data. His earlier reports following the initial release of ChatGPT placed social work educators in the top 20 jobs with the greatest exposure. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4375268
Great resources. Thank you. Looking forward to diving into both of these.