How bad is the problem of unreliable answers from gen AI systems? A month-long study by the BBC using ChatGPT, Microsoft Copilot, Google Gemini and Perplexity determined that 51% of answers to queries had “significant issues of some form.” Almost 20% of answers based on BBC content had factual errors while 13% of quotes were altered or just didn’t exist.
A district judge has ruled that, as a matter of law, Ross Intelligence’s use of Thompson Reuters’ Westlaw content to train its own legal research model is not “fair use” but copyright infringement. This is a revision of a 2023 ruling and leaves issues for a jury to decide. It might, however, shut down one line of defense when it comes to AI scraping of copyright material.
Adoption of gen AI has slowed as marketers adjust to its limits, according to a study by Deloitte Digital. Use of gen AI to create newsletter and advertisements, automate content creation and produce graphic content is down, while the number of respondents using gen AI to create short-form text or brainstorm content ideas is up YoY (7% and 19% respectively).
It barely feels like news when industry giants add AI features, since that’s what they’re expected to do. But lots of people use their products, so their announcements are probably of higher interest than most. With that in mind, let’s note that Adobe has just announced general availability of agents for audience creation, journey orchestration, experimentation, journey insights, website optimization and agent support.