deepseek doesn't need to outperform other models, it just needs to be cheap, or, efficient
the cost of deepseek (if it's true) will disrupt the logic of current AI industry
The current AI industry is built on a financing bubble, where investors hand over money blindly without demanding that companies profit from AI.
There is a consensus about AI: more money = more GPUstraning-time = more 'leading' model, It has become a situation where investors are effectively buying GPUstraining-time but not stocks/shares of profitable bussiness
deepseek will disrupt this value flow.
> Alibaba Cloud announced the third round of price cuts for its large models this year, with the visual understanding models of the General Qwen-VL models experiencing a price reduction of over 80% across the board. The Qwen-VL-Plus model saw a direct price drop of 81%, with the input cost being only 0.0015 yuan per thousand tokens, setting a record for the lowest price across the network. The higher-performance Qwen-VL-Max model was reduced to 0.003 yuan per thousand tokens, with a significant decrease of 85%. According to the latest prices, one yuan can process up to approximately 600 720P images or 1700 480P images.
the cost of deepseek (if it's true) will disrupt the logic of current AI industry
The current AI industry is built on a financing bubble, where investors hand over money blindly without demanding that companies profit from AI. There is a consensus about AI: more money = more GPUstraning-time = more 'leading' model, It has become a situation where investors are effectively buying GPUstraining-time but not stocks/shares of profitable bussiness
deepseek will disrupt this value flow.
> Alibaba Cloud announced the third round of price cuts for its large models this year, with the visual understanding models of the General Qwen-VL models experiencing a price reduction of over 80% across the board. The Qwen-VL-Plus model saw a direct price drop of 81%, with the input cost being only 0.0015 yuan per thousand tokens, setting a record for the lowest price across the network. The higher-performance Qwen-VL-Max model was reduced to 0.003 yuan per thousand tokens, with a significant decrease of 85%. According to the latest prices, one yuan can process up to approximately 600 720P images or 1700 480P images.