I. The ChatGPT Phenomenon and Issues
User Retention Problem
Hundreds of millions have tried ChatGPT, but most have not returned. It quickly entered the public eye, but many people only use it occasionally and have not formed a habit of continuous use. This may be because it takes time for people's habits and ways of thinking about new tools to change, or it may be because LLMs are not products themselves and need to be restructured in terms of framework, user experience, and tools to become useful.
Comparison with Other Products
Products like e-commerce and the iPhone took a long time to be widely accepted and used, while ChatGPT, despite rapid user growth, has a low retention rate. It has leveraged the infrastructure built over the past 25 years, allowing a large number of users to try it quickly, but it still requires more effort to be truly accepted.
II. Enterprise Application Situation
Usage Proportion
Surveys show that enterprises have a great interest in LLMs, but the deployment rate is low. The adoption rate varies greatly across different departments, such as being more useful in coding and marketing, and less useful in legal and human resources.
Pilot and Deployment Differences
Many enterprises have only conducted pilot projects, with fewer actually applying them to their business. For example, although Accenture has many projects, most are pilots rather than full-scale deployments.
III. Limitations of LLMs and Development Directions
Limitations
LLMs appear to be products, but they are not. Applying them directly to areas like search poses problems, such as Microsoft's unsuccessful attempt to combine them with Bing, because the output of LLMs may be inaccurate, and specialized products need to be built around them.
Development Directions
The models may improve to some extent, such as adding agents, voice, and multimodal functions to solve more problems. At the same time, it is necessary to transform them from technology to actual products, and meet user needs through restructuring.
IV. Investment and Market Trends
Investment Situation
A large amount of investment has poured into the LLM field, such as the growth of NVIDIA's data center revenue, and the platform capital expenditure of hyperscale enterprises may also increase. This investment boom may have skipped the slow and painful phase of product-market fit, and there is a risk of forming a bubble.
Market View
One view is that this is an investment bubble, and another view is that these startups are collectively betting that LLMs are a technology that needs to go through the traditional customer discovery process to achieve product-market fit, rather than considering LLMs themselves as products. In the end, LLMs may go through a process of development and improvement like the Internet, e-commerce, and the iPhone, rather than changing everything immediately.