No in-house AI, no comment

The Financial Times has launched a generative algorithm chatbot, Ask FT, based on Anthropic’s Claude and trained with several decades of its own articles. The move highlights the logic of generative AI: data must put it to good use, either by selling it to a third party (as is the case of Reddit in the United States, Prisa in Spain or Le Monde in France) or by using it to create algorithms capable of producing viable services.

In the case of the Financial Times, which has a prestigious 135-year archive of economic, financial and political content, it makes perfect sense to use this content to answer users’ questions. The choice of Claude is also no accident: from the outset, the company has focused on making its algorithm very efficient at ingesting and processing large amounts of text.

Needless to say, people have had fun with it, trying to trick the algorithm, and sharing the results on social networks. All generative algorithms are prone to hallucinations, the result of low correlations that do not represent real relationships, but which nevertheless the algorithm will still try to see a connection with, and that effect is magnified when many users go all out to “fool” it.