AI-sprint experiment

I think most of us have had really fun testing out AI-tools since the LLM-revolution started followed by an abundance of tools and use cases.

Although our Slack channels are filled with takeaways interesting experiments, we found that it has been a barrier including it in our daily work.

Many at SVT are justifiably wary of safety issues and have high quality standards which has been impeding experimentation. Building trust for tools is a slow process while losing it is instantaneous.

Cross-organizational AI-sprint

To lower the barriers we tried out a cross-organizational AI-sprint a few weeks ago. We’ve for a long had our tech sprints described here as a great way to try out wild ideas and new tech. What if we did the same with an AI-focus? We set off with five teams with developers, analysts, editors, content experts, designers etc

The intent was simply to learn as much as possible in a week. The setup was work with whatever tool, with whatever and whoever you like.

Except for creating the time and cleaning calendars for the sprint we also tried enabling it by helping out with tools, crippling bureaucracy and a relaxed security policy.

What did we do?

A week later we had a demo with prototypes ranging from tech framework comparison to end user focused features in SVT Play.

The Alastor team wanted to add mood tags for different shows, describing feeling, tone and style of shows by analysing video content and metadata like subtitles etc.

They ended up with often amazingly useful and accurate mood tag suggestions for editors to use.

One team focused on building a conversational interface for analysis using Slack as the most convenient interface. Build on GPT-4o+RAG, the Slackbot #prompt-pal was born

The most user-facing team was AiDA, creating a conversation AI interface in SVT Play making an exploring conversation to find things to watch

Two people from the Growth Marketing team (without developers) built a copy assistant for summarizing title descriptions for push notifications (and started using it in production the next day).

Now what?

In a follow-up survey all but one said that we should do it again. The last person thought that we should include it in our daily work instead. Insightful!

Some comments were “The funniest week since I started at SVT” and “It made AI a bit less magical or rabbit hole nerdy…”

So we’ll probably continue in some form, hopefully with even broader participation!