Trump called Sam

[Deep See Shock]

It was only a few days ago that Trump called Sam Altman and Son Jeong-ui to invest 100 billion dollars + 400 billion dollars in AI infrastructure in the U.S. A total of 500 billion dollars. The whole world was noisy. At the same time, Microsoft and Meta each announced that they would invest more than 50 billion dollars each in AI infrastructure this year. The so-called infinite competition for AI infrastructure seemed to begin in earnest.

By the way, DeepSeek, a Chinese AI startup, quietly launched its V3 model on Jan. 20. It's open-source-based, so anyone can verify it and process it as they please. What's amazing is its cost-effectiveness.

With a $5.8 million investment in just three months, the open-source model's performance matches that of OpenAI's GPT-4o. API fees are around 5% of OpenAI's. In addition, DeepSeek can handle the latest search + inference requests indefinitely. With performance above the same, DeepSeek can ask unlimited queries, though it is limited to three per day in Perplexity Pro mode.

I've developed new algorithms and models, and the performance is amazing. It works well in Korean, too. I don't think I need to use Perplexity more because I have excellent reasoning, analysis, and Q&A skills.

To do domain AI, you have to train with an open-source AI model with only internal data from the company or organization, and I don't think the cost will be too much of a burden at this cost-effectiveness.

If it is true that this performance is achieved without high-performance GPUs, the winner of the future AI war is China. While the U.S. is struggling with Tran's war and stock price stimulus, Chinese startups are devising all their wisdom and innovating at a breakneck pace. I'm afraid to imagine what the future will be like if these high-performance lightweight AI models continue to be improved and are embedded in Chinese robots.

I recommend you try it first.