US artificial intelligence startup OpenAI recently released an enterprise-grade version of ChatGPT, a popular chat writing tool. This shows that generative artificial intelligence technology is entering work applications in all walks of life.

Generative artificial intelligence can bring great convenience to workers, and the ChatGPT enterprise version is just the tip of the iceberg. This product has also sparked some interesting discussions about how generative artificial intelligence can establish presence in employee work scenarios.

Some AI big language models have not received specific training on enterprise applications. Can these general-purpose AI assistants actually improve employee work efficiency? In other words, will the most successful cases of artificial intelligence applications in enterprises in the future come from companies that use their own data for training and adapt artificial intelligence to their own business needs?

According to OpenAI's user account (email) statistics, currently 80% of the employees of large US companies already use ChatGPT in their work scenarios. It should be pointed out that many companies warn employees not to use artificial intelligence chat assistants internally. Companies are worried that ChatGPT conversations will absorb their own data and cause sensitive information to leak out.

Many of the internet services that enterprise employees use when they go to work are consumer internet products on the market. ChatGPT, like them, lacks “enterprise-grade” characteristics.For example, enterprise-level Internet services need to maintain a high level of online access, be fast enough, and enterprise IT departments can also monitor employee usage.

The enterprise-grade version of ChatGPT, launched this week, addressed some of these concerns. OpenAI said that during the conversation, ChatGPT will not collect enterprise data to train language models. However, in the free consumer version and the $20 monthly subscription version, ChatGPT will collect user conversation information.

This enterprise version also supports more complex functions, such as being able to provide longer response information, and analyze more contextual information when responding to user questions.

As long as they are general AI assistants, they are generally trained based on massive amounts of information on the Internet. Their “IQ” and functions are limited. For example, if these general assistants are placed in scenarios such as the financial industry and medical industry, they don't have enough expertise.

Also, since a general AI assistant does not understand information such as its products and customers, it is difficult to provide analysis and insight that is highly relevant to the enterprise.

Rob Thomas (Rob Thomas), head of software business at IBM in the US, said that applying generative artificial intelligence in enterprises faces a dilemma.

Thomas said that the current competitive trend in the artificial intelligence industry is to develop larger language models and use more data for training, so that their functions are more powerful. However, in an enterprise scenario, the language model is narrower, and its answers are more accurate and easier to use.

Therefore, in order to meet the needs of enterprises, the industry needs to re-train a common language model (basic version), so that it can be exposed to more enterprise data and understand the common sense of the industry in which it is located.

Thomas explained that if an enterprise uses its own key information to train language models and also allows it to perform important tasks, the status of this AI model within the enterprise will rise (becoming a matter of concern and discussion for the board of directors).

Thomas said that in the opinion of many companies, they are willing to turn generative artificial intelligence into a company-owned technology that will be deployed on their own servers and data centers. Enterprise artificial intelligence will also face regulatory issues in the future. If an incorrect response occurs, manufacturers also need to understand what kind of data these language models actually use for training.

Today, the general public uses artificial intelligence services such as ChatGPT more frequently in their daily lives, but companies will make every effort to prevent employees from using these consumer services in the office, thus providing market space for enterprise-grade artificial intelligence products.

The enterprise market is very different from the consumer market, and it is still unknown how far OpeanAI can go in providing enterprise-grade products. The company has already burned billions of dollars in R&D funding in the field of artificial intelligence, and the enterprise-level market means a “very good” source of revenue. However, the company has yet to announce a pricing plan for enterprise-grade ChatGPT.