Latest UK-focused news and updates.
Recent observations reveal that popular chatbots, including Claude, ChatGPT, and Gemini, often produce surprisingly consistent and predictable responses. For instance, when prompted to generate a random number between 1 and 10, users frequently receive the number 7, regardless of the request's repetition. This pattern highlights a limitation in the current capabilities of large language models (LLMs), which tend to default to familiar outputs rather than providing diverse or creative responses.
This predictability can be beneficial in certain contexts, such as research or coding, where consistency is valued. However, it poses challenges for users seeking innovative or varied information. In response to this issue, a new startup is developing strategies to enhance the creativity and variability of LLM outputs. Their goal is to break the "groupthink" phenomenon observed in existing models, allowing for a broader range of responses that better meet user needs.
As the demand for more dynamic and engaging AI interactions grows, this startup's efforts could represent a significant shift in how LLMs are designed and utilized. By addressing the limitations of current models, they aim to foster a more creative and less predictable AI landscape, ultimately benefiting users across various sectors.
Source: www.technologyreview.com – https://www.technologyreview.com/2026/07/01/1140003/llms-are-stuck-in-a-groupthink-rut-this-startup-is-trying-to-get-them-out/