AI Applied: Covering AI News, Interviews and Tools - ChatGPT, Midjourney, Gemini, OpenAI, Anthropic DeepSeek: The Rising Star of AI
Jan 16, 2025
DeepSeek, a rising AI model from a Chinese company, is turning heads with its cost-effective development. The hosts dive into how it stacks up against giants like GPT-4, showcasing its innovative features and performance capabilities. They also confront the troubling ethical implications surrounding censorship and data sourcing, raising questions about its training methods. Amidst heavy restrictions, DeepSeek still shines, offering potential privacy advantages and a fresh perspective on the evolving AI landscape.
AI Snips
Chapters
Transcript
Episode notes
DeepSeek's Development
- DeepSeek, a Chinese AI company, developed a GPT-4 competitor.
- Notably, they achieved this with only $5 million in two months, despite sanctions limiting access to top-tier chips.
DeepSeek V3 vs. GPT-4 and Claude
- A Reddit user tested DeepSeek V3 against GPT-4 and Claude, finding it superior in reasoning and math.
- However, DeepSeek's output was suspiciously similar to GPT-4, suggesting potential training data overlap.
DeepSeek's Censorship
- Conor tested DeepSeek's censorship by asking about the Chinese government's flaws.
- The model initially answered but then deleted its response, claiming it was beyond its scope, while readily criticizing the US government.