Ancestors

Toot

Written by joshuatj on 2024-11-18 at 22:41

First time at #ALGIM24 (Annual conference of the Association of Local Government Information Management). Huge conference with 500 delegates. Six parallel streams, spanning ICT, web/digital/comms, customer experience, geospatial, Smart Communities and information management, with collaboration as the focus.

(Personally I don't like parallel streams, but I can understand it's a great way to attract sponsors/vendors.)

[#]ALGIM2024

=> More informations about this toot | More toots from joshuatj@digipres.club

Descendants

Written by joshuatj on 2024-11-18 at 22:41

Keynote was by @audreyt and I just have to share the beautiful poem-prayer:

When we see “internet of things”, let’s make it an internet of beings.

When we see “virtual reality”, let’s make it a shared reality.

When we see “machine learning”, let’s make it collaborative learning.

When we see “user experience”, let’s make it about human experience.

When we hear “the singularity is near”, let us remember: the plurality is here.

[#]ALGIM24 #ALGIM2024

=> More informations about this toot | More toots from joshuatj@digipres.club

Written by joshuatj on 2024-11-18 at 23:01

IT Award Finalist: Nelson City Council worked with AI Factory to streamline analysis of public submissions to its long term plan. OpenAI secure endpoint was used so that data doesn't get retained and used for further model training. The software interface designed for easy lift and shift for use by other councils. Total budget $20,000. Inspiring.

[#]ALGIM2024 #ALGIM24

=> More informations about this toot | More toots from joshuatj@digipres.club

Written by joshuatj on 2024-11-18 at 23:14

IM project finalist: Dunedin City Council - AI-driven classification for Information Management. Scope: Building Warrant of Fitness (BWOF) compliance. Frustrated by inefficiencies.

Objective: All relevant documents in one folder.

4 months. Constant iteration. Had conversation with Archives New Zealand. Hope to get more external collaboration for auto-classification. See attached for project tech. (1/🧵)

[#]ALGIM2024 #ALGIM24

=> View attached media

=> More informations about this toot | More toots from joshuatj@digipres.club

Written by joshuatj on 2024-11-18 at 23:22

LLM for Classification (ChatGPT 3.5 Turbo). Lots of prompt engineering. Lots and lots of iteration to get the prompt right.

Trialled llama, gemini, on local GPUs, just couldn't get the accuracy they need.

Total cost depends on the total number of tokens. Discovered more is not always better.

QA key part of the detailed solution. Sampling. Might result is our first EDRMS disposal (Yayyy!)

(2/🧵)

[#]ALGIM2024 #ALGIM24

=> View attached media

=> More informations about this toot | More toots from joshuatj@digipres.club

Written by joshuatj on 2024-11-18 at 23:32

93% accuracy. Wild success. Est savings of 160 hours per annum (1 mth of FTE)

Azure dev and prod cost: ~$332 NZD

QA cost: ~$30 NZD

Actual production run: ~$150 NZD

💪💪🤩😮😮

Learnings: corruption detection, metadata enhancement, composite documents, summaries

(3/🧵)

[#]ALGIM2024 #ALGIM24

=> View attached media

=> More informations about this toot | More toots from joshuatj@digipres.club

Written by joshuatj on 2024-11-18 at 23:44

Super excited about their next steps. Collaboration being the highlight. Bright future for auto-classification using LLM. 🔆 🌱🌼🌻

(4/4)

[#]ALGIM2024 #ALGIM24

=> View attached media

=> More informations about this toot | More toots from joshuatj@digipres.club

Written by Michael U on 2024-11-19 at 03:20

@joshuatj “set a new benchmark in government records management”, says the programme 🤔

very interesting though, thanks for sharing

=> More informations about this toot | More toots from jet@cloudisland.nz