Bitman on Nostr: For those unfamiliar with DeepSeek 👇 China has developed a rival model to GPT-4 at ...
For those unfamiliar with DeepSeek 👇
China has developed a rival model to GPT-4 at 10% of the cost, using limited Nvidia chips (H800) and open-source code.
Their AI model cost $5.6M (vs. $100M for GPT-4), utilizing Nvidia H800 chips and an MoE (Mixture of Experts) architecture.
This could be a “thorn in the side” for Nvidia, as it demonstrates that the company’s stock may have been overvalued based on the assumption that the AI revolution would require much more advanced chips than are actually necessary.
Essentially, they’ve reached OpenAI’s level at roughly 27x less cost than OpenAI’s LLM o1.
This macro movement is turning the market into a “one trade play,” increasing asset correlation with a repricing of big tech valuations.
This shift isn’t directly related to #Bitcoin, so those who can distinguish FUD from structural change will benefit from better positioning.
China has developed a rival model to GPT-4 at 10% of the cost, using limited Nvidia chips (H800) and open-source code.
Their AI model cost $5.6M (vs. $100M for GPT-4), utilizing Nvidia H800 chips and an MoE (Mixture of Experts) architecture.
This could be a “thorn in the side” for Nvidia, as it demonstrates that the company’s stock may have been overvalued based on the assumption that the AI revolution would require much more advanced chips than are actually necessary.
Essentially, they’ve reached OpenAI’s level at roughly 27x less cost than OpenAI’s LLM o1.
This macro movement is turning the market into a “one trade play,” increasing asset correlation with a repricing of big tech valuations.
This shift isn’t directly related to #Bitcoin, so those who can distinguish FUD from structural change will benefit from better positioning.