Skip to content
Peak Text: AI and the Golden Age of Libraries and Archives by Keir Winesmith
- Finished “EGOT of GLAMs” with latest job
- Mapping Brisbane
- 1957 Tram network: based on older tracks, evolved into suburbs. River is fixed
- Averaged with AI = River + Tramlines
- Maps of Queensland
- Merged many maps of Queensland with Model that knows birds.
- NFSA (National Film and Sound Archive), Machine Learning and AI
- Pilot to have AI transcribe etc material in the archive
- Internal Transparency
- Principals of NFSA AI project
- Maintain Trust – Train only on stuff they have copyright
- Build effectively and Transparently
- Create Public Value
- AI = Archival Intelligence
- or maybe “Average Inputs”
- Stereograms created by AI
- Defaults to the small subset that is online
- Previously was 1900 colonial pictures
- Now still colonial but Google products a Sanfran street scene
- The perfect Training Data is what archives have been putting out for years with lots of metadata
- OpenAi Whisper trained on lots of youtube videos it turns mumbles into “Like and subscribe” and music fade outs turn into “Than you for watching”
- The new golden age
- Previous Golden Age was films explosion between the wars
- 1980s and 1990s of Video games
- Australian stories are no longer being made on celleloid and now being on social media
- Thinks as boomers die off Facebook is dying off.
- Other platforms my die in the next few years
- New sites just algorithmically created content, not stuff shared by friends etc
- What does NFSA do in response to how things change
- Ability to search transcripts mean they can find people taking about something or someone, not just titles
- Mass Transcript + Graph. References to cultural things like movies, quotes in unrelated documents.
- Transcribed 18.7 years of content
- Hope to open up more later in 2026
- But don’t forget openness got us in this mess in the first place
- Need to think before publishing stuff, since now it will be ingested by everyone
The Evolution of the OCI Artifact Revolution by Andrew Block
- Modern Eras of Computing
- What technologies came out of the cloud native era – Containers
- The power of containers
- Resource Management
- Consistency
- Speed
- The Container format wars – docker vs rkt
- Docker Ecosystem tied closely to Docker Inc
- The Open Container Initiative
- Image Spec, Runtime Spec, Distribution Spec
- “Containers are just fancy files and fancy processes”
- Image Manifest
- Just a json file
- Media Type header will come up later
- Expanding beyond Container images
- OCI can store Artifacts which are content types other than container images
- Registry must explicitly support it (most of them do now)
- New stuff you can store
- Signature
- Software packages ( .jar, rpm )
- OCI Image and Distribution Spec 1.1
- Released 2024
- artifactType or mediaType
- Can refer to other artifacts (ie signature for container) and API supports both directions to discover
- Benefits of OCI Artifacts
- Standard
- Centralised Management
- Reuse existing tools
- Evolve existing practices
- What Projects use it
- Helm and Homebrew both use it.
- Notary, Sigstore, etc use it to store signations etc of other Containers
- Argo CD and Flux CD store manifests within OCI artifacts. Easier to give prod servers access to OCI registry rather than git repo
- Kubernetes OCI Image Volume – Not exactly a OCI Artifact
- Tooling
- skopeo and crane let us inspect OCI metadata
- ORAS – Create and manipulate OCI artifacts
- The Evolution of the OCI Artifact Revolution by Andrew Block
- AI
- Currently uses git, hugging face, Object Storage to store stuff
- Challenges. Several types of content, lack of standards ways to store and use
- ModelPack is potential standard solution
- Leverages stuff already in OCI
- Demo with helm (using report software called “zot”)
- Can push chart to oci: url
- ORAS
- ORAS can push a simple artifact . Even a simple plain text file
README: The Developer’s forgotten love letter by Swapnil Ogale
- Technical Writer at AWS
- “Customers will jump straight to the README, not to your comprehensive docs” – A Senior Developer
- Story about how a powerful tool with no documentation doesn’t get any traction. A better documented tool that is less powerful gets more traction.
- It is the first impression of your product. Sometimes the only impression
- Anatomy of a good README
- The Hook
- Getting Started
- Examples
- Beyond the Basics
- Building Trust
- The Hook
- Start with user’s pain point, not your technical achievement
- Problem Solver not Technical Jargon
- Getting Started
- What do I install, what version, command that wroks, One good example, where to get help
- Beyond the Basics
- Full Docs, How to contribute
- Building Trust
- License information
- Maybe Contributor list
- Readme driven development
- Design for users first
- Think like a user
- The User Journey
- What is this?
- Will it solve my problem?
- Can I try it easily?
- What if I get stuck?
- The first 30 seconds
- What makes them stay
- Clear problem statement
- Easy setup instructions
- One problem example
- What works for users?
- Write like explaining to a friend
- Use Screenshots and gifs when helpful
- Break up walls of text
- Test on fresh machine
- update when things change
- What frustrates users – anti-patterns
- “It is easy, just”
- Assuming I know the jargon
- “See the source for details”
- Installation steps that don’t work
- No examples
- Some Templates and Tools
- AI Tools
- Loses personality
- Make sure it has examples
- Has example AI prompt and wrapper script that we will share
- Key Takeaways
- Users are not lazy, they’re busy solving problems
- “Obvious” is not obvious to them
- Examples > Explanations
- Test instructions ohttp://joinbookwyrm.com/n real users
- README Maintenance is feature work