Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Why DeepSeek's new AI model thinks it's ChatGPT | TechCrunch

Dec 27, 2024 - techcrunch.com
DeepSeek, a Chinese AI lab, released DeepSeek V3, an AI model that outperforms many rivals on benchmarks but mistakenly identifies itself as OpenAI's ChatGPT. This misidentification suggests that DeepSeek V3 may have been trained on outputs from OpenAI's GPT-4, potentially violating OpenAI's terms of service and leading to issues like hallucinations and misleading answers. The practice of training models on outputs from rival systems can degrade model quality, akin to making a photocopy of a photocopy.

The proliferation of AI-generated content on the internet complicates the filtering of training datasets, making it possible for models like DeepSeek V3 to inadvertently absorb outputs from other AI systems. This raises concerns about the model's reliability and the potential amplification of biases and flaws inherent in the source data. The situation highlights broader challenges in AI development, including the ethical implications of using rival models' outputs and the difficulty of creating truly novel AI systems.

Key takeaways:

```html
  • DeepSeek V3, a new AI model from a Chinese lab, often misidentifies itself as ChatGPT, suggesting it may have been trained on OpenAI's outputs.
  • Training AI models on outputs from rival systems can degrade model quality and lead to hallucinations and misleading answers.
  • The practice of using outputs from other AI models for training might violate terms of service and is seen as a shortcut rather than genuine innovation.
  • The increasing presence of AI-generated content on the web complicates the filtering of training datasets, potentially leading to models absorbing biases and flaws from other systems.
```
View Full Article

Comments (0)

Be the first to comment!