news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

No matter how successful an AI startup is, it has not found PMF.

Author | Li Shiyun
edit| Jingyu

Character.AI, one of the most watched unicorns in the field of artificial intelligence, has ushered in a new destiny. Although recent media reports said that Elon Musk's xAI was considering acquiring Character.AI, it was Google that finally reaped the fruits.

On August 2, Google announced that it had brought Character.AI’s core team (including founders Noam Shazeer, Daniel De Freitas, etc.) under its wing, and signed a cooperation agreement with Character.AI (including licensing its model) and providing financial support to the company.

By absorbing the core team and funding the original company, the Silicon Valley giants have completed the "acquisition" of technology startups in disguise.It cleverly avoids regulation but takes away the core assets of startups.

This seems to be the common destination for artificial intelligence startups in Silicon Valley.

At the end of last year, when OpenAI was in a "palace fight", Microsoft tried to recruit the core team including Sam Altman, but failed. However, in March this year, it used this method to win Inflection, and the core team including CEO Mustafa Suleyman joined Microsoft. In June, Amazon also won Adept through similar means.

So, as the first in the AI ​​emotional companionship track and a super app on par with ChatGPT, why did Character.AI fail to avoid this fate? Why is it still difficult for it to develop independently?

Behind the change in fate of Character.AI, what common problems are hidden in current artificial intelligence startups?

01

CHaracter.AI is difficult to develop independently

s reason:There is TPF, but no PMF

At present, AI super applications mainly focus on two directions: one is productivity, represented by ChatGPT, and the other is virtual social interaction, represented by Character.AI.

Founded in 2022, Character.AI allows users to create any virtual characters and talk to them. For example, Daenerys, the "Dragon Mother" in "Game of Thrones", Musk, and even the game characters in "Genshin Impact". It essentially satisfies the user's fantasy and provides emotional and companionship value.

Judging from user data, Character.AI is a very successful application: in just two years, its daily active users have reached nearly 6 million; user time and stickiness are also very high - within 5 months of the product's launch, users sent more than 2 billion messages, and active users used it for more than 2 hours a day.

However, applications are just superficial, and the company's ultimate vision is to achieve AGI.

The two founders, Noam Shazeer and Daniel De Freitas, were both top technical talents at Google. Noam Shazeer is one of the authors of the core paper on the Transformer architecture, "Attention is all your need", while Daniel was responsible for Google's large language model project LaMDA.

They also represent the typical thinking of technology entrepreneurs: they believe in the integration of production and modeling, that is, developing the underlying large model while developing upper-level applications, improving the application experience through the underlying technology, promoting the technology through applications, and ultimately feeding back to the underlying model technology - forming a positive cycle.

Based on this idea, Character.AI has continuously upgraded the underlying model capabilities, including launching a virtual character call function that matches the GPT-4o voice interaction. From the perspective of technology and product performance, Character.AI has achieved TPF (technology product matching) in a certain sense.

Character.AI's virtual character voice call function | Image source: Character.AI

However, the core problem is that users accept the value of this product.，But not willing to pay for it.

Currently, the company's main income comes from membership subscriptions, which allow users to get faster response speeds, more virtual characters, etc. However, at present, Character.AI has less than 100,000 subscribers, accounting for less than one thousandth of the total number of users.

Character.AI also seems to have noticed this problem. Since March this year, it has started recruiting professional product and operation teams, but the impact has been minimal.

An investor in the AI ​​field once told Geek Park that it would be more difficult to charge for AI products. Taking the Internet era as an example, most To C applications are free and ultimately make money through advertising, search, and recommendations. For Character.AI, users use the product for a clear purpose, which is "emotional value". Users have a very low acceptance of advertising, and the product's design is not suitable for the advertising model, which makes it difficult to find a reasonable business closed loop.

In the past two years, the Character.AI team has been working hard to reduce the product's inference cost, which has dropped by 33 times since its establishment. But in the end, it is still difficult to break even in terms of input-output ratio. It can be said thatIt forms TPF but not PMF (user-market match).

This is also a common problem in AI applications today. Even OpenAI is facing a loss of nearly $5 billion this year, according to media reports.

Ultimately, like other AI startups, they need to rely on external funding to survive. After receiving $150 million in funding from Andreessen Horowitz (A16Z) last year, Character.AI's financing road does not seem to be smooth. In September last year, foreign media reported that it was negotiating with Google for a new round of financing with a valuation of more than $5 billion, but the investment has not yet materialized.

As Musk vehemently denies his intention to acquire Character.AI, Google may be the company's best option.

02

Each in its place:

The founding team went to a large company to pursue AGI.

The rest of the people try to find PMF

From this perspective, this transaction shows that Character.AI has finally woken up. The grand dream of forming a closed business loop through production and model integration and gradually supporting the company's progress towards AGI is no longer feasible.So everyone returned to their respective positions and made more pragmatic choices.

For the founding team, joining the former company with strong financial resources is more likely to support their AGI dream. For Character.AI, the core proposition is how to make good use of the user base and increase commercial value.

Character.AI’s predicament does not mean that the emotional companionship track has no user value, but that it has not maximized its commercial value. It may need to introduce better product and operation talents, think about better product design and matching business models, and truly make AI applications profitable.

Before this path is feasible, the company chooses to reduce costs and increase efficiency first.In addition to this team change, Character.AI also announced that it would abandon pre-trained models and switch to open source models. This is essentially to reduce training costs, use resources in more important places, and allow the company to break even on PMF faster.

Many AI applications have a "platform dream": they believe that the current commercial value can be put aside and the most important thing is to increase the number of users. However, the story of Character.AI shows that this is quite difficult.

If products in the mobile Internet era have network effects, for example, for TikTok, when there are more users, there will be more content, and more content can eventually be generalized to more users, ultimately creating a huge content platform.

But for AI products like Character.AI, when more and more users chat with AI, it cannot fundamentally change the content supply and form a positive cycle of content and consumption. Under this product definition, it is difficult to take the platform path.

Virtual characters on the Character.AI platform | Image source: Character.AI

The story of Character.AI and other AI startups being "swallowed" by giants may indicate this truth:

1) It is not enough to only achieve TPF. If users spend time but do not pay, and there is no way to convert it into other commercial income, then PMF has not been found. It is equally important whether technology can derive products that attract attention, and whether good products can derive good commercial value.

2) Good product managers and product innovation will be extremely important. Just as Steve Jobs said, "Technology alone is not enough." Great technical talents must be equipped with great product talents. In addition to the logic of technology, there is also the "logic of the human world" that must be understood in order to achieve true product innovation.

Character.AI may have figured out these two points, but chose to give up its persistence and efforts on them, which is related to the pursuit and characteristics of the founder. But for other artificial intelligence startups, this exploration story may still have a lot to do.

*Header image source: Visual China

This article is an original article from Geek Park. For reprinting, please contact Geek Jun on WeChat: geekparkGO

Geek Question

What do you think of the AI ​​emotional companionship track?

Why is it so difficult to develop independently?

Musk: I slept on the factory floor for three years. When employees changed shifts, they would be more responsible when they saw me.

Like and followGeek Park Video Account，