Chinese AI startup DeepSeek has emerged, Meta employee reveals: training costs are lower than our supervisors' salaries!

A Chinese AI startup called "DeepSeek" has emerged, releasing a new artificial intelligence model called DeepSeek V3 on the second day of Christmas. This month, another inference model, DeepSeeker R1, has been open sourced again. The industry has evaluated that it has approached or even surpassed OpenAI's o1 model in some fields, which has brought tremendous pressure to Silicon Valley's tech giants.

Recently, a Meta employee complained on the anonymous workplace community TeamBlind about Deepseek (a Chinese AI company)'s breakthrough in the field of artificial intelligence (AI). Deepseek V3 released Llama 4 in benchmark Shanghai, which brought huge pressure to Meta.

The employee said:

To make matters worse, this is actually an 'unknown Chinese company' that has achieved such a level with only $5.5 million in training budget. Now, engineers are crazily dismantling Deepseek, trying to copy anything that can be copied from it - I'm really not exaggerating at all

The employee further revealed that the management of Meta's Gen AI department is currently quite anxious because the operating costs of the Gen AI department are too high compared to Deepseek's training costs of only about $5.5 million.

The management is very concerned now because the operating costs of the Gen AI department are too high. How can we explain to the top management? Even more exaggeratedly, every so-called 'leader' in our department earns a salary higher than the total cost of training Deepseek V3, and we have dozens of such 'leaders' here

In addition, an anonymous employee from Google also commented under the article that what DeepSeek is doing is indeed crazy. In addition to Meta, OpenAI, Google, and Anthropic are also very nervous now, but such competition may also accelerate the growth of AI technology.