• For consultant David Caswell, they are also extending the scope of journalism, for example, by facilitating the creation of scripts with specifically adapted language from print articles for audiences that prefer audio content to the written word.
• News organisations are aware of the growing importance of large language models in the media landscape and are investing in staff training in a bid to optimize their use of the new technology.
How should journalists respond to the development of AI in their profession?
Using AI has become an essential skill, and it is one that is a bit like driving or riding a bike, you don’t learn about it through reading but by doing it. That’s why all journalists should now be exploring the possibilities offered by AI. If they don’t learn how to prompt properly, it will compromise content they obtain from AI. On the other hand, if they master the use of these tools, they will have control over the results they generate which will improve the quality of their articles. More and more major news organizations are teaching prompting, which they see as critically important, to large numbers of staff. Some of them are investing in training even though they don’t yet know how their journalists will use it. There is a simple reason for this: regardless of how AI develops, they know it will be a big part of their day-to-day work. Media organizations therefore want to have the flexibility to adapt to new tools. Today, a lot of newsrooms are developing their own AI systems, which is relatively easy to do. Often these are tools that enable journalists to execute prompts just by clicking on a button.
La tendance générale pour les rédactions est d’avoir davantage de contrôle sur les modèles de langues
Why is there so much emphasis on large language models?
Large language models are a bit like electricity. They are general technologies which are not designed for a particular purpose but can be used for a huge range of tasks. And with regard to LLMs we don’t yet know what many of these tasks will be. Today these tools are widely used in a lot of newsrooms, but only in a shallow way — that is to say that they have yet to result in a great deal of observable change. Their influence on workflows and especially on the production of published content is still relatively low. For example, AI is often used for summarizing articles, which doesn’t have much impact on the production of content. But that is a typical use case right now. So too are copyediting, search engine optimization, adapting texts for different audiences, and generating social media posts.
More and more major tech companies and start-ups are developing tools for newsrooms…
Start-ups like Lede.ai and Scroll.ai provide tools to produce translations and summaries etc., which make life easier for journalists who only need to import an article to instantly obtain results. Major companies are developing more comprehensive systems, notably Google with Google Genesis which includes news gathering and workflow management components.
The general trend is for newsrooms to take more control over LLMs, and Nordic countries leading the way on this. At the Guardian, they have also fine-tuned an existing model to adapt it to their needs. Today technologies like retrieval-augmented generation (RAG) can enable LLMs to produce results that are more relevant and include recent contextual data: these models don’t just respond on the basis of their training, they can also read live data and include it in their answers.
There has been an increase in the availability of audio versions of articles that have been processed by AI. Is this a major trend?
Today it’s almost impossible to distinguish between real and synthesized voices. Some tools, like Synthesia.io, even allow you to control emotion in voices. However, the subtlety of text-to-speech is that articles which are written to be read are not always suitable for listening audiences. This is why some news organizations like Agência Publica in Brazil have created AI tools to transform print articles into audio scripts. Other media companies like Channel 1 in the United States are planning to use AI generated anchors to present round the clock news. You might wonder whether the quality is good enough, but the reality is that this type of content will appeal to audiences that are happy with this level of quality.
David Caswell
