A busy full week back started with some activity on my ‘personal researcher notes’ investigation, which has already received quite a few notes in a variety of formats from the researchers who are looking into the BT Archives’ files. GPT is very adept at reading handwriting now, so I am finding that it is easy to work with notes in any form that the researchers send them. The question is what prompt garners the best response, so I am running a trial and error process to produce a prompt that can be universal.

I have also started an interesting discussion with BT about what it would take for them to trust the descriptions enough to place them in the catalogue, which I think will form part of a larger discussion about trust in digital tools. BT have also been creating their own automatically generated descriptions using Whisper, so this is already an area they are prepared to tackle.

It was great to see so many people at the Mining Review workshop on Tuesday, a huge thank you to Daniel and Patrick for pulling that together. For my own work, it prompted a useful conversation about RAG which has now continued between Nayomi and myself, looking at how to apply the technique to the GPO Circulars. I’m very excited to have recruited Nayomi to join the fight to get the circulars into a query-bot format, starting with the 1979 PDF!

In the meantime, I have OCR’d and tidied ten years of Bradford-specific information in the circulars from 1880 – 1889 and am ready to put these through the MyGPT process. I am looking at the way this sort of tool could be used for linkage, whether this means feeding it additional relevant data from other sources (eg. The Postal Museum Archive), creating a query bot that focuses on a specific type of data (staff newsletters), or simply using the MyGPT as a jumping off point for linkage-speared research (searching for the answers to questions such as ‘what technology does this circular mention, and which museum has it?’ or ‘what happened in Newcastle at the same time?’).

For computer-vision and gender: I only have one day left of LLaVA running in the background until the entire dataset is done, and Anna-Maria and I are in fruitful discussions with the team at Stepstone about using their gender bias decoder before the end of the month.

In terms of outreach: Anna-Maria and I have put in an abstract focused on the gender work for the ECREA conference this coming September, and I sent off a pitch to The Conversation on the theme of ‘historians using AI’ as promised in the fellows catch up. Next week I am going to investigate the potential to have a computer vision cluster-specific article in Museum Crush, which I am happy to talk about in more detail to anyone who has interest in this article or another potential one!