Search results for Google Gemini 2.5
AI & Enterprise
AI models hallucinate 28 percent of software dependency upgrade recommendations, study finds
Leaving software dependency decisions to AI models can be risky, a study found. DevSecOps firm Sonatype analysed 36,870 dependency upgrade recommendations generated across four package repositories and reviewed 258,000 recommendations produced by seven AI models from Anthropic, OpenAI and Google. In an earlier study, nearly 28 percent of upgrades recommended by OpenAI GPT-5 were hallucinations involving non-existent versions or paths. A follow-up study found errors persisted in newer models.
AI & Enterprise
Overuse of AI in writing makes it bland and produces sentences humans would never write
A study found that excessive use of large language models can change not only writing style but also the content itself. Researchers on the U.S. West Coast ran an experiment with 100 participants on how money affects happiness and compared responses by reliance on LLMs. Heavy users submitted more neutral answers and wrote in a less personal, more formal style. They rated their writing as less creative, though satisfaction was similar, a point researchers flagged as a long-term concern.
AI & Enterprise
Meta delays next-generation AI \'Avocado\' launch until after May due to performance issues
Meta has delayed the launch of its next-generation AI model, Avocado, from this month until after May, the New York Times reported. The delay follows internal performance tests that showed the model trailing leading offerings from Google, OpenAI and Anthropic. The report said Avocado fell short of the latest rival models in reasoning, coding and writing evaluations. Meta executives are also discussing a temporary licensing option for Google’s Gemini for its AI products, but no final decision has been made.
