The future of large language models and perspectives on investment
Machine-assisted translation of the archived article. Read the Chinese original.

With regard to GPT-4, we wonder why it is developing so quickly. We would like to know how much of its model improvements are and how many parameters it uses. OpenAI no longer publishes this information, although they were a non-profit organization. Former GPT-2 was even open source, but since Microsoft investment, OpenAI has become a private company. Because of the high level of competition in the market, this information has become a business secret and we cannot be informed. I can't even tell you what the parameters are.
The article surveys competitors to OpenAI’s GPT-4, including Google’s BERT, Baidu’s ERNIE Bot, Alibaba’s Qwen, Meta’s LLaMA and models from Huawei. It also mentions AI platforms founded by former employees of established companies.
In fact, these bottom-up models all seek universality, just as little kids do in general education. However, if you want to train an artificial intelligence for a particular field of expertise, you can choose to train yourself, such as a big company like the financial giant Bloomberg. But for small and medium-sized companies, they may not have that power.
Smaller organisations can adapt a pretrained foundation model to a specialist field or call a model through its API. The author says this was the starting point for MyGPTLife.
The table compares the GPT models available through MyGPTLife and their API charges at publication.
Model name | Output price | Rate limit | Token limit, including context |
gpt-3.5-turbo | $0.002 / k | 90K / min | 4K |
gpt-3.5-turbo-0613 | $0.002 / k | 90K / min | 4K |
gpt-3.5-turbo-16k | $0.004 / k | 180K / min | 16K |
gpt-3.5-turbo-16k-0613 | $0.004 / k | 180K / min | 16k |
gpt-4 | $0.06 / k | 200K / min | 8K |
gpt-4-0613 | $0.06/ k | 200K / min | 8K |
gpt-4-32k | $0.12 / k | 150K / min | 32K |
gpt-4-32k-0613 | $0.12 / k | 150K / min | 32K |
