@hugh
That was a fantastic write-up! Thank you!
Recent searches
Search options
#ff2024
0 posts0 participants0 posts today
This and #FF2024 in general make me think we need updated studies on information seeking in different types of GLAM collections. And that all modern searches need a 'semantic search off' toggle to enable exact matches again
Aus GLAM SpaceHugh (@hugh@ausglam.space)This isn't what I expected to publish as my write-up of #FF2024 but I guess that's what happens when you give yourself a month to process things:
The fantastic futures of Slow AI and vibes-based search
https://www.hughrundle.net/fantastic-futures-and-slow-ai
What a delight it was to present at the #ff2024 #ff24 Fantastic Futures #conference last week, held at the #NFSA, celebrating all things #AI and #ML in the #GLAM sector.
See what the #ANU School of #Cybernetics got up to at the conference:
A great example of learning on the job from Asa - 'Making cursive comfortable: Transcribing 19th Century archives using AI models' https://prov.vic.gov.au/about-us/our-blog/making-cursive-comfortable as presented at #FF2024 #AI4LAM
prov.vic.gov.auMaking cursive comfortable | PROVUsing Transkribus With over 100 shelf kilometres of historic Victorian government archives, to both digitise and transcribe, from eye-bending cursive to readable text, Public Record Office Victoria has relied on volunteers to diligently perform these tasks so family and history research can be made possible online. While typeface has been relatively easy to transcribe by automation tools (e.g. Trove) handwriting has been more difficult to auto-transcribe with any accuracy.
As if prompted by #FF2024, RNZ ran this article featuring Peter Lucas-Jones and the great 92% accuracy Te Reo Māori text-to-speech work at Te Hiku Media.
"It is important to recognise that big tech is not going to save te reo Māori."
https://www.rnz.co.nz/news/alert-nat/531303/why-google-maps-can-t-pronounce-maori-place-names
RNZ · Why Google Maps can't pronounce Māori place names
Continued thread
#AI4ALAM folk, you outdid yourselves at #FF2024! Your optimism, your cynicism, your questions about how we get the AI / ML we deserve - and most importantly your support for each other. Please keep sharing and learning together! https://ai4lam.org
ai4lam.orgai4lamAI4LAM is an international, participatory community focused on advancing the use of artificial intelligence in, for and by libraries, archives and museums.
@nfitzger announces that #ff2025 will be in London 3-5th December 2025 at the #BritishLibrary
#AI4LAM #ff2024
Continued thread
We hope you enjoyed #FF2024! Save the date for #FF2025 Fantastic Futures 2025 AI4LAM at the British Library, London 3-5 December 2025 https://ai4lam.org
Continued thread
Kara Kennedy, AI Literacy Institute – AI Literacy for Librarians in Low-Resource, Multicultural Communities
Two choices, don't use AI (like early days of Wikipedia) or learn how to use AI wisely
1. Engage AI champion
2. Leverage existing content - LinkedIn learning. Lots of free resources.
3. Create accessible resources
Librarians, GLAM, we can lead AI literacy, help prepare community for AI
Continued thread
Continued thread
End up, created our own - Whisper Scribe. https://www.github.com/gsu-library/whisper-scribe (Rachel invited us to check it out)
Future development: partitions, dynamic user permissions, locking records when editing, investigate syncing existing transcript, project profiles.
Continued thread
Rachel Senese Myers – Adopting Whisper: Creating a front end optimized for processing needs
Georgia State University
Oral history. Ohms. Not searchable.
Experiment: faster Whisper. Took learnings from #FF2023. Blew Adobe Premiere Pro out of the water. IT said, you can't do this - command line. Need UI.
Whisper Web UI.
No database. Browser crash, data gone.
Whishper. Love database. No batch upload. No diarization. No batch edits.
Continued thread
Lightning talks (5 mins each).
Asa Letourneau – C19th Handwriting and Model Making with Transkribus
Public Record Office Victoria, State Archives Victoria
Rapid learning and experimenting. Started July 2023. Tried Trankribus. Not successful. Created 3 private model, CER 5%. Tested public model Text Titan, 2.75% CER.
Now Trankribus beta site. Recognise tabular record. Hoping to deploy IIIF Glycerine for crowdsourcing editing of transcription.
Continued thread
Replied in thread
Human to judge generated similarity. 900 judgements by 15 NLA staff. Head to head comparison. Individually, CLIP highest score. Combination of model considered best 66% of the time against other variants. Open source models getting better.
Continued thread
Nice comparison tool from NLA to see which models did better at text search results and image similarity searches #FF2024
