Again, some investigation specific updates from me are below. For the gender investigation, I recommend checking out Anna-Maria’s latest update which covered everything 😊

Beyond investigating, I was pleased this week to be invited to share my researchers notes work in a TaNC webinar in February.

Personal researcher notes

I have seven active researchers on the go now, with a few more waiting to be set up. I’ve deliberately asked for a range of note formats, ranging from handwritten, to typed, to structured within a template I provided.

One researcher has already returned some notes to me, and I have turned these into descriptions using GPT. I would welcome any thoughts on these! I think in the future I will ask GPT to focus on the fact the notes it has are from an archival file, and ignore terms such as “archive” and “collection,” as well as ask GPT to avoid emotive summary statements.

BT Archives POST/33/4983 (circa 1936-1938)

This collection comprises internal memos, draft letters, and copies of original correspondence related to British Telecommunications' (BT) K6 “Jubilee Kiosk” in North Riding and Westmorland. The archive is divided into two sections. The first addresses general objections to the red colour of the kiosks, including discussions within BT and responses to external parties. The second section, from 1938 onwards, focuses on specific requests for colour changes or alternatives to the K6 kiosk in particular villages. Key correspondents include Captain then Major T. L. Dugdale M.P., Hurbert G Thornley, and Catherine McDougall of the CPRE.

BT Archives POST/33/3809B: Objections to Standard Colours (1930-1949)

This collection contains correspondence and documents regarding objections to the standard red colour of BT kiosks. It includes discussions with various entities over the preferred colours for K1, K2, and K3 telephone kiosks in different locales. Key correspondences include requests for colour changes to green, black and white, and other hues to match local aesthetics or historical precedents. Notable interactions involve the MCC at Lords Cricket Ground, The Institute of Electrical Engineers, various local councils and corporations, and private landowners. The archive captures the evolution of colour preferences and the compromises made, illustrating the tension between national standards and local preferences in the design and placement of BT kiosks during the early 20th century.

GPO Circulars Work

I decided to play with MyGPT while I wait to progress with the GPT API. MyGPT has quite a tight limit on how much information you can feed into the knowledge base, so I have started with two small experiments and will see how close to feeding it all 100 years I can get in the new year.

My, rather untechnical, pipeline has involved running a circular through ABBYY FineReader to create a searchable document. I have then searched for all instances of the word ‘Bradford’ in this document. I then exported entire sections where Bradford was mentioned, so if Bradford came up as one telephone exchange in a section about telephone exchange openings I exported all openings, and created one large RTF file including nothing but references to Bradford in any given year.

My first test was with the 1979 Gazette, which is the most recent circular we have and therefore easiest to get a clean OCR reading from. I did not edit any information I exported from ABBYY at all, and asked GPT to create a query bot that would provide answers based only on the source it was given and say “I don’t know” in situations where the question was not answerable from the source content.

Anyone with a GPT4 subscription can play with the MyGPT themselves at this link – if you do, please let me know your thoughts! For those without, I’ve stuck a few screenshots of interactions with the bot below to show the results.

For my next MyGPT experiment, I turned to the 1883 Circular as this is the oldest we have. My process was much the same as working with the 1979 Gazette, although I would take the time to tweak ABBYY’s RTF output if it was too complicated for GPT to understand. This meant teaching ABBYY that some strangely formatted lists were actually tables, and asking it to ignore marks on the scan that it thought were letters. I tried to be as light touch as possible, especially as GPT seems to do quite well when dealing with unpleasant formatting that even human readers might struggle to understand.

The 1883 MyGPT (which you can play with here) starts to answer some telegraphy questions we have had, and I think shows that there is promise in the earlier circulars. I am now working on the 1887 – 1889 circulars, which are quite involved as there is more need to tweak on the OCR, but if just a few days of tweaking could mean we have a GPT able to understand 100 years of GPO involvement in Bradford, I think it would be worth it!