DeepSeek has released an updated AI model, R1-0528, which demonstrates significant performance on math and coding benchmarks. However, there's controversy regarding its training data, as some researchers suggest links to Google's Gemini AI family. Melbourne developer Sam Paech presented findings indicating that DeepSeek's phrasing resembles Gemini's outputs, leading to speculation about the ethical practices in AI training. This is not the first instance of DeepSeek facing scrutiny; previous models have been accused of using data from competitors, including claims of training on ChatGPT logs and the use of distillation to extract data from larger models.
DeepSeek's R1 model has reportedly been trained using outputs from Google's Gemini, raising concerns about the ethical practices in AI data sourcing.
Evidence suggests that DeepSeek's latest AI models show a preference for phrasing similar to that of Google’s Gemini, indicating potential data sourcing issues.
Collection
[
|
...
]