Instead of building tools on top of open-source models, publishers are increasingly taking the route of signing deals with generative AI companies to access their proprietary models — although they need to negotiate terms to have some control over the data they are sharing.
– Sara Guaglione
Digiday surveys which style of large language models are more appropriate for journalism. Open-source models are available with no licensing or upfront costs, while proprietary models, such as OpenAI‘s ChatGPT, control access from the owner, usually at a cost.
Those favouring the proprietary models say they offer superior performance, industry-leading development teams, and support for customization.
Advocates of the open-source models believe they provide more opportunities for bespoke applications, and their wide user base means more people are fine-tuning the model, for example, improving bias and security.
Some journalism companies have begun customizing large language models to take advantage of their in-house editorial data and know-how.
The case for and against open-source large language models for use in newsrooms | DIGIDAY | August 14, 2023 | by Sara Guaglione