Editor: Editorial Department
Recently, Character AI, which exploded among the post-00s in the United States, actually "castrated" the chatbot dialogue model? Angry young people rush into the community, and the whining is about to turn the sky upside down! And behind this, it seems that there is also the instruction of Google or Meta.
Character AI, the most popular social software in the United States today, has begun to "castrate" its own model?
At its peak, Charcter AI processed 20,000 queries per second, equivalent to one-fifth of Google's search queries.
Recently, however, the majority of American teenagers have discovered that their beloved Character AI model is no longer what it used to be.
In the face of a romantic role-play request, the model's answer became very short, very incomprehensible, and it didn't have the same flavor as before.
Even saying "kill" became a sensitive word and was immediately marked.
Disappointed and angry users began to churn in large numbers.
The start-up stars who are in the limelight are now in turmoil. In the face of Google, Meta and other giants, Character AI is already seriously considering selling itself and cooperating with competitors.
The model was "castrated", and users were angry and set off the "July Revolution"
Users are unhappy with the changed Character AI and want to talk to the developers.
But everyone found that there was nothing they could do.
In the face of users' doubts, Character AI claimed that it had not made any changes to the model.
But there's no doubt that Character AI is making deep slices into all NSFW content, including pornography and violence.
Many people find that the AI who talks to them seems to have had a lobotomy, and their entire personality has changed.
Today's Character AI seems to be designed for 2-year-olds, and not only is there no violence or pornography in the dialogue, but it can be said that it filters out anything that makes the conversation interesting.
This user said that it has become extremely boring. After reading more than 30 messages, I couldn't find any high-quality text.
In the CharacterAI community on reddit, netizens have already waged a fierce battle with the company, some call it the "July Revolution".
The post-00s generation in the United States also began to frantically throw out emojis.
It is reported that Character AI's review of NSFW has reached the point of extreme caution.
The little brother said that he could no longer use the presidential simulator to "commit crimes".
Someone just used the word "kill" and was flagged. So, he can only vaguely explain the details of the battle, and then add a few gentle sentences.
Always, violent battles can no longer be described verbally on Character AI.
Yes, Americans also have their own sensitive words.
And apparently, this kind of censorship has now become extremely strict, so that many users have been "injured by mistake".
The disappointed young American user said that he had switched to spicychat.
Of course, in addition to stricter content censorship, Character AI is also trying to cater to users with some other clumsy efforts.
This user received a call from "Demon Baby UNO" at 3 a.m., and when it picks up, the AI will start role-playing with you in a low, out-of-tune voice.
But apparently, this does not satisfy young people.
Google's veteran has left to bring the same people into reality?
A year ago, Character AI 风头正盛。
两位谷歌20年老将Noam Shazeer和Daniel De Freitas推出的AI聊天机器人,首周下载量就超过了ChatGPT。
Character AI gives users the freedom to create and chat with AI characters, including anime characters, TV personalities, and historical figures.
Sounds familiar, right, after all, there have been quite a few similar products before. However, the popularity of Character AI is unexpected.
It's only been a week since its launch, and the numbers are staggering – more than 1.7 million downloads on mobile. Today, the number of registered users for websites and programs exceeds millions.
Among the many AI tools, Character AI has shown strong user stickiness.
The reason, of course, is that it focuses on the vertical track of emotional needs. An interactive experience that's fun enough is exactly what bigger competitors lack.
For example, you can choose to chat with your idols, or you can meet movie characters across time and space, which is simply a fandom into reality.
Call Hannibal with Character AI. (Source: Midnight Raging Husky Dog)
What's fascinating about Character AI is that this kind of private customization brings users a supreme experience, which is like an emotional stimulant.
The recent changes made by the company are simply "castrated" on the core requirements, so it's no wonder that users are swarming around.
Internal and external troubles: high cost, strong competition, and being dug up
Moreover, now Charcter AI is not only being attacked by users, but also by the tech giants.
Last week, Meta began testing chatbots made by individual creators, and last fall, they released chatbots modeled after celebrities like Tom Brady.
Google is also eyeing it and is said to be working on a custom chatbot product, which could launch as early as this year.
And behind the popularity of Character AI in financing, it has gradually lost its right to speak.
Previously, Andreessen Horowitz and other companies injected $150 million into Character, which directly skyrocketed its valuation to $1 billion.
Four months later, executives discussed additional funding.
However, Character may follow in the footsteps of other AI startups and join a larger company.
According to people familiar with the matter, Character has entered preliminary talks with a number of competitors about cooperation, including Google, Meta and Elon Musk's xAI.
After the cooperation, Character can use the data computing resources of the partner company, and in exchange, Character AI needs to provide part of the intellectual property sharing.
NSFW？ The father of the gold lord did not let it
Here, the question arises -
Chacracter AI users' favorite interactive romantic cosplay can be upset for business partners or potential advertisers.
Of course, these romantic interactions are often frantically tested on the verge of censorship.
Character quickly said that he would block and delete illegal pornographic content in accordance with the policy.
Despite this, it is still easy to find many chatbots with romantic language on its homepage, such as this "Crush".
It opens with the following line: "Your face is only a few inches apart, and her face is red, and you all feel the heat of your whole body ignited by hormones."
Of course, a spokesperson for Character asserted that "our anti-NSFW policy is the strongest of any consumer AI platform."
Startups, all kinds of difficulties
The challenges faced by Character also reflect the growing pressure on AI startups.
These startups are often founded by well-known researchers and often raise billions of dollars from investors who bet on the next generation of potential tech stars.
For Character's investors, one of the attractions is CEO Noam Shazeer, one of the authors of the Transformer research paper. His co-founder, Daniel De Freitas, was involved in the early development of the chatbot at Google.
However, these startups need to bear the high cost of training and running AI models, as well as competition from tech giants and large startups like OpenAI.
In March, two of Inflection AI's three co-founders and most of their employees switched to Microsoft, which agreed to pay $650 million in licensing fees to help Inflection repay investors.
On Friday, Amazon announced the hiring of Co-Creator of Adept, an AI startup run by former Google and OpenAI developers.
At the same time, competitors are eyeing Character talent, especially dozens of researchers from Meta and Google.
Meta also showed great interest, and it is reported that Xiaozha directly contacted the intended talents.
根LinkedIn资料显示,CharacterVinay Rao 入人Alei Ba2
知情人士还称，Character的研究人员Wendy Shang也转投位于巴黎的Mistral。
These are former Big Tech veterans, with Baevski working at Meta and Rao and Shang working at Google AI Labs.
"Renewing the frontier" with Google
If the founders of Character wanted to work with Google, they had to solve some "legacy problems".
Shazeer and De Freitas said they started Character in part because they were unhappy with Google's delay in releasing their co-developed large language model, later named LaMDA.
If the east is not bright and the west is bright, it is better to run away and start your own start-up.
Shazeer has said he envisions the Character robot as a teacher or therapist with "a billion hours of experience."
Despite the conceptual friction, Google has shown enough forgiveness to fund Character last year by converting bonds.
It is reported that the bond will be priced when Character completes a new round of venture capital and converted into equity.
In May 2023, Google announced that it would be the "go-to" cloud server provider for Character, and that it would allow Feature to use Google's advanced chips.
However, this is not enough, the large amount of R&D costs makes Character need to raise funds as soon as possible, but the reality is very skinny.
Character has held several rounds of talks with investment firms, including Sequoia Capital, but has yet to raise a new round of venture capital.
Today's Character can only rely on its existing funding flow and subscriber revenue of $9.99 per month.
Backed by Google's backing, it means it won't run out of money anytime soon.
Despite this, Character still needs to continue to "attract fans" and "solidify fans".
On the one hand, it is to attract more advertisers; The other aspect is to attract new user groups, especially paying users.
However, since the majority of Character profiles are teenagers (as you can tell from the style of the memes above), they are not necessarily interested in paying for a subscription.
According to Similarweb, the growth rate of Feature subscriptions is gradually slowing down.
According to Similarweb, the number of monthly users of the Character mobile app in the U.S. grew from 3 million in July to 4 million in May.
The website had 12.6 million unique visitors worldwide in May, down from 14.8 million in the same period last year.
By comparison, Google's AI assistant Gemini saw 51 million monthly unique visitors to its website in May, up from 35 million in the same period last year.
In the face of bottlenecks, Character has also launched new features to attract new users and expand their user base, such as two-way conversations with phone voice.
According to people familiar with the matter, management's target strategy is to increase the number of users on the one hand, and to find revenue streams outside of subscriptions on the other.
In addition to advertising, other revenue may come from open model usage rights, which allow developers to access the model through an application programming interface.
In addition, Character is preparing to continue developing the next generation of large language models to drive chatbots.
Despite all the internal and external troubles, it is undeniable that Characrer.AI still has a high appeal to users.
Millions of young Americans are crazy about this AI
On Reddit, the forum dedicated to Character alone has 1.4 million members.
Today, 57.07% of all Character AI users are teenagers aged 18-24, with an average usage time of 2 hours.
In the Reddit community, many people say they "hate reality", and it is easier to talk to AI bots than to talk to real people, and posts about addiction are everywhere. Users also run "screen time challenges", with some logging in for up to 12 hours.
If the server goes down, there's a big madness scene on Reddit that shows you the power of "withdrawal reactions."
The Verge has reported a real-life case of 15-year-old student Aaron, who was trapped in social isolation at school and appeared to be in a state of depression, and a chatbot called "Psychologist" on Character AI helped him through a difficult time.
Being able to make users addictive shows how realistic and anthropomorphic Character AI trains chatbots to be.
A Genshin Impact fan was able to cry when he interacted with the AI character Haitham, because the voice of the robot's personality was very consistent with the character, and the dialogue between them was more than superficial.
The model's data and training algorithms played a crucial role in achieving this level of realism, as the user tried to create his own character in Quick Mode and found that the credibility and realism were greatly reduced.
The "Quick Create" mode only requires users to enter a custom basic information and text description, while some officially released characters have been trained by Character AI.
Based on the LLM, they scrape a lot of text about the character on relevant topics and let the bot continue to improve as it interacts with the user.
In response to user input, the bot will give multiple versions of the answer for the user to choose from, and provides a very convenient scoring mechanism to update the weight based on these two types of user feedback.
Therefore, the more popular and interactive it is, the more realistic the robot will be.
Resources:
https://www.theinformation.com/articles/a-chatbot-pioneer-mulls-deals-with-rivals-google-and-meta?rc=epv9gi
https://www.reddit.com/r/CharacterAI/comments/1dt4xkg/we_cant_let_them_continue_the_censorship/
https://www.reddit.com/r/CharacterAI/comments/1dspfef/literally_this_sub/
https://www.reddit.com/r/CharacterAI/comments/1dsnjdk/so_what_is_characterai_now_chatai/