How to do magic with the power of ‘scoped’ AI in online publishing – Labrador CMS

In an average newsroom, only 5% of reporters have ChatGPT4. Only some know how to use it. So, what happens if you introduce scale? And give 2 500 journalists ChatGPT4?

Generative AI can turbocharge online publishing, a Norwegian startup is proving – but only when it restricts AI capabilities.

Jan Thoresen, CEO of Labrador CMS, showed the 100-strong audience at a recent Mx3 AI event in London how the startup added generative AI into its toolkit to create and publish digital publications and magazines quickly and easily.

He emphasised, however, that Labrador CMS restricted AI’s capabilities to work exclusively with the facts in submitted stories, not allowing it to access external information or facts beyond the provided content. This limitation controls the AI behaviour and aligns it with the intended purpose so that it addresses specific needs without venturing into unrestricted or undesired functionalities, often referred to as the scoped use of Generative Pre-trained Transformers (GPT). 

While all onstage discussions at the London event were off-the-record, Jan shared some insights in an interview afterwards. He explained how AI integration uses tools such as GPT for tasks related to organising, optimising and streamlining content workflows rather than creating content. GPT enables Labrador CMS to generate subtitles, optimise articles for search engines, and add tags for enhanced searchability. It also creates catchy headlines for social media platforms – all in real time. 

Scope to do your magic

These tasks happen within a defined and limited scope, confined to specific rules, guidelines, or boundaries. The scope determines the range of tasks, information, or functionalities the AI model can perform or access. 

“So that’s one of the advantages of having it built into the system,” Jan said. “It’s not a free space where you can play around with GPT. This is a restricted area, so we restrict GPT not to use any facts or information besides the body text that we submit. So, only use this text and do your magic with it.”

At the outset, this involves significant amounts of prompt engineering, but once set up is “very easy to use”.

At the day-long event on Hoxton Square in Shoreditch, Labrador CMS also launched its fully automated AI translation function. It included a demonstration of the AI automation of article translation into 141 possible languages, encompassing the automated inclusion of metadata, fact boxes, image captions, titles and tags. According to Labrador CMS, this capability opens up new business opportunities for cross-language publishing and will help publishers improve content distribution. Read more here.

A comprehensive set of tools

Serving 260 publishers with more than 2 500 journalists in ten countries, Jan explained how Labrador CMS recognises the diverse needs of its clients. Because each publisher  – not only newspapers and magazines but also B2B publications and TV stations – has unique requirements, the shared platform ensures access to a comprehensive set of tools. These tools encompass the sum of all the individual needs and ideas contributed to by thousands of journalists. Feedback from users fosters ongoing improvements that shape the platform’s evolution.

Jan said he envisions further improvements to the startup’s Software as a Service (SaaS) platform. This includes automated A/B testing and optimisation of teasers, as well as plans to expand and improve image creation, recognising the potential to provide journalists with better graphics.

* Watch the interview with Jan here.