I may be wrong, but #LLM stochastic parrots don't seem to be a useful tool for a research scholar in the #humanities. I can't see it doing anything novel or labour saving for me. Maybe write a first draft for an abstract? Which takes no time for me and only happens like five times a year?
@mrundkvist I think they are very useful, at least if one uses them for the tasks in which a language model can contribute.
I will give you the example of a project I finished today. In a historical archive they are several teen-thousands of free text records written in English and Hebrew. The researchers need to get insights of all the people that appear there: biographic data, live places, membership to organizations, etc. To extract the data: people, places, organizations, dates, and relationships between then such as Person A belong to the Org A
Person B was deported to Place B in Date B
etc, with a height performance, the language models are very useful.
What they cannot do is to replace the researcher, of course.
@mrundkvist Depends of the entity type and relation that we extract.
Let us say that for all data the accuracy is 86%. In some cases such as membership of a person to an organization, the accuracy was higher than when humans were looking for it.
As you see, the result doesn't replace the role of the researcher, as was often claimed by media or interpreted by people. It is just a good tool to provide data for research, preprocess it, etc.
Add comment