News publishers have disputed a claim fromGooglethat using their content to “fine-tune” its AI models contains “no realistic prospect of harm” to them.
‘Fine-tuning’enables an AI model that has already been trained to add new data to respond to a specific prompt or task.
Google told the UK’s Competition and Markets Authority that there is “no realistic prospect of harm to publishers in respect of training/fine-tuning ofAImodels for search and search generative AI features.
“Fine-tuning helps the model learn how to process information rather than what current information to display; this internal processing does not create a substitute for publisher websites. A model relying solely on patterns learned during training would be static, often outdated, and prone to hallucinations.”
The CMA has said it is not minded to provide separate controls for fine-tuning but publishers asked it to reconsider.
The CMA has published responses from several major news organisations to its consultation relating to proposed conduct requirements to be imposed on Google under new powers inthe Digital Markets, Competition and Consumers Act 2024.
Guardian Media Groupsaid in response to Google’s position that it “demonstrates the clear value of publisher content at every stage: training, fine-tuning, and grounding – but in particular in keeping outputs accurate.
“Given that news publishers are trusted sources across a wide range of queries, this would suggest that their content is being used to create substitutional products – invalidating the argument that it causes ‘no harm’ to publishers.”
The Guardian argued there is a “trend towardsmore frequent, bespoke fine-tuning to improve model accuracy, the impact of which is to improve the recency of underlying LLMs, i.e. their ability to include current information in their responses.
“This in turn risks devaluation of emerging and valuable RAG markets for news content.”
Source: Press Gazette
