<aside>
💬
"Go into the archive for yourself, rather than blindly trusting the AI output."
</aside>
Wednesday October 28, 2026 · 7:25 PM · Opening night
Jacob Ortiz and Daniel Shavit are winners of the Internet Archive Builders Fellowship.
Key ideas from the talk
Jake: showing the gaps
- He built a tool on Internet Archive Canada's digitized government publications, starting with public health.
- It shows where records are missing, where jurisdictions are thin, and how language shifted over time. A confident AI answer shows none of that.
- Every answer links to the original document. When the evidence is thin, the tool says so.
- About 45% of the pilot's documents had no date.
- He used care principles for records that show historical harms, and wants to work with the affected communities next.
Danny: the city's decisions
- He pulled a hundred years of San Francisco council records into a map: who voted, who got paid, where the money went.
- With help from the Internet Archive team, he got Vancouver data too.
- A small model you run at home can give precise answers when it queries a clean, real dataset. He calls this model sovereignty.
Quotes
- "Historical records, they're not straightforward, so they shouldn't deserve a synthesized, collapsed answer." Jacob Ortiz
- "This tool is supposed to be sort of an introduction for you to explore and not replace your own critical thinking and your own human judgment." Jacob Ortiz
- "AI can produce a statement that's true when properly cited, but it could answer the wrong questions." Jacob Ortiz
- "You can actually see the whole history of an issue, all the decisions that were taken in the past, and all the people who were connected to these decisions." Daniel Shavit
- "This has enabled people to participate better in their democracy and have more transparency about what's going on around them." Daniel Shavit