DeepSeek may have used Google's Gemini to train its latest model
a year ago
- #DeepSeek
- #AI
- #Gemini
- DeepSeek released an updated R1 reasoning AI model performing well on math and coding benchmarks.
- Evidence suggests DeepSeek's R1-0528 model may have been trained on outputs from Google’s Gemini.
- DeepSeek has previous accusations of training on rival AI models like OpenAI’s ChatGPT.
- OpenAI suspects DeepSeek used distillation to train its models, violating OpenAI’s terms of service.
- AI-generated content flooding the web makes it hard to filter AI outputs from training datasets.
- AI companies like OpenAI and Google are increasing security to prevent data distillation.
- OpenAI now requires ID verification for advanced model access, excluding China.
- Google and Anthropic are summarizing model traces to protect competitive advantages.