<ul><li>Chinese lab DeepSeek released an updated version of its R1 reasoning AI model last week, speculated to have been trained on data from Google's Gemini family of AI.</li><li>Melbourne-based developer Sam Paech claims evidence suggests DeepSeek's R1-0528 model was trained on outputs from Gemini, as observed through similarities in word preferences.</li><li>DeepSeek has faced accusations before, including training on data from rival AI models like OpenAI's ChatGPT. Distillation techniques have been linked to DeepSeek, raising concerns about violating OpenAI's terms of service.</li><li>AI companies are increasingly enhancing security measures to prevent distillation, with actions like ID verification for model access and steps to summarize generated model traces to protect competitive advantages.</li></ul>

DeepSeek may have used Google’s Gemini to train its latest model

Discover more