news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

New Intelligence Report

Editor: Aeneas is so sleepy

【New Wisdom Introduction】Oh no, the CEO and president just ran away? Character.AI's CEO left Google with 30 employees. This is the third "CEO" runaway incident in just 5 months. Big boss's comment: The AI ​​bubble will burst in a few weeks.

Another AI startup CEO has run away!

Just now, the AI ​​community was shocked by this news——

Character.AI CEO Naom Shazeer, who left Google to start his own company, left the company with President Daniel De Freitas and a large number of members of the research team and returned to his old employer Google!

Character.AI will license its LLM to Google in exchange for more funds

The company's general counsel, Dominic Perella, will serve as interim CEO.

The departing CEO also took away Character.AI’s employees responsible for model training and voice AI, which is 30 of the 130 employees.

They will join Google to participate in the Gemini AI project.

The remaining 100 people, in their hearts, be like——

The deal reportedly values ​​CharacterAI at $2.5 billion, nearly half of the $5 billion it discussed with investors last year.

After the last Inflection CEO left the company and joined Microsoft, this is another startup CEO who ran away in a gorgeous way.

What we mean is, it is no longer popular to acquire companies, but to directly acquire the founders?

In fact, as early as a month ago, the precarious business situation of Character.AI had already begun to emerge.

Although at its peak, Character.AI's performance was dazzling (the number of mobile users reached 4 million, the website's monthly visitors reached 14.8 million, and at one point reached 1/5 of Google's search queries).

Now, as expected, after half a lifetime of wandering, I finally returned to Google.

Google welcomes defecting employees with open arms

Character.AI officials explained their decision to "sell themselves" as follows:

When Noam and Daniel founded Character.AI, our goal was to achieve personalized superintelligence, which required a full-stack approach. We had to pre-train models and perform post-training to support Character.AI's unique experiences while building a product platform that could reach users globally. However, the landscape has changed over the past two years; more pre-trained models are now available. Given these changes, we see an advantage in making greater use of third-party LLMs for use in conjunction with our own models. This allows us to invest more resources in post-training and creating new product experiences for our growing user base.

In November 2021, Google employees Noam Shazeer and Daniel De Freitas left their former employer because they were dissatisfied with Google's bureaucracy.

Previously, Shazeer led the team to build LaMDA, and the incident caused by this caused a sensation.

After leaving, the two founded Character.AI, one as CEO and the other as president.

Three years later, the old employer opened its arms and generously welcomed the return of the "defected" former employee.

Google agreed to pay Character.AI a licensing fee for its model and to hire the CEO and several researchers at a high price.

The CEO said he was very excited to return to Google and become a member of the Google DeepMind team.

And the Character.AI business that I left behind will get better and better in the future.

I'm proud of what we've accomplished at Character.AI over the past three years. I'm confident that the funding from the non-exclusive licensing agreement with Google, combined with the amazing Character.AI team, will enable Character.AI to continue to be successful in the future.

Google said Shazeer would join the DeepMind research team, but did not specify his or De Freitas' exact roles.

We are very happy to welcome back Noam, a distinguished researcher in the field of machine learning, who will join the Google DeepMind research team along with a small number of colleagues.

Of course, Shazeer will also resign from Character.AI's board of directors, while a16z partner Sarah Wang will continue to serve as an independent director.

Character.AI gets a big boost

Moreover, Character.AI’s leaders also informed employees that previous investors would receive a buyback at $88 per share.

This is about 2.5 times the value per share of Character.AI in its 2023 Series A round, when the company was valued at $1 billion. (It had previously raised a total of $193 million.)

At the same time, Google also promised that it would provide more funding to Character.AI to help it continue to grow and continue to build personalized AI products for global users.

It is reported that Character.AI will turn to open source models such as Meta's Llama 3.1 to support its products rather than its internal models.

Over the past five months, Silicon Valley has given us a glimpse into the “acquisition of founders” craze.

Whether it is Character.AI, Adept, or Inflection, their transactions can allow investors to quickly recover their funds.

However, The Information said this was far from the huge returns that venture capitalists expected when they funded them.

As for unvested employee stock options, they will continue to vest at $88 per share until the end of July 2026.

Character.AI guarantees that investors will continue to get paid through a fund paid out of the licensing agreement. Options will continue to vest after two years, but there will no longer be guaranteed payouts from the fund.

Although the deal is not strictly considered an acquisition, the operation has attracted the attention of antitrust regulators.

Now, the Federal Trade Commission has begun investigating Microsoft's deal with Inflection to determine whether it should be considered an acquisition that Microsoft should report to the government.

AI chatbots, a bleak future?

But now, behind the prosperity, silence has returned, and everyone has noticed this fact: training conversational AI models is extremely costly, but the number of users willing to pay for it is far from enough to recoup the cost.

In this critical moment, Character.AI has been negotiating with several large technology companies for several months.

The negotiating parties also include Zuckerberg's Meta Platforms and Musk's xAI.

Previously, Google had provided Character.AI with cloud computing services and access to its advanced chips, and provided financing through convertible notes.

In addition to the cost issue, another major reason for AI chatbot revenue has also been blocked.

Character.AI has become an instant hit among young people across the U.S. because it allows users to communicate with personalized chatbots, including anime characters, TV celebrities, and historical figures.

Of course, we all know what the maximum capacity of chatbots is.

Character.AI has previously raised more than $150 million from venture capitalists including a16z and Greycroft.

However, since they have received money from the sponsor, their business development will not be so free - the interactive romantic cosplay that Chacracter AI users love most may make business partners or potential advertisers dissatisfied.

Character.AI has stated that it will regulate content more strictly and all explicit content will be blocked and deleted.

Forced to castrate their own models, they will have even fewer paying users.

After discovering that the model had been castrated, angry American young people directly launched the "July Revolution" in the "CharacterAI" community on Reddit.

These AIs have been "castrated" so severely that they have become a group of "useless people"

Just a few days ago, news broke that Meta will stop using celebrity AI chatbots.

This also proves that chatbots are not currently a good implementation direction for the popular AI models.

NYU professor: AI bubble will burst in a few weeks!

In this regard, NYU professor and best-selling author Marcus said he had foreseen this.

In his opinion, it’s time to burst the bubble of generative AI.

Inflection’s CEO left the company and joined Microsoft. Character.AI’s CEO and president both fled and joined Google. Now, only Stability AI is left struggling.

Moreover, Microsoft's CFO just said that it may take 15 years for the company to recoup its investment.

Isn't this going downhill?

If 2023 is the year of AI’s promise, then 2024 is the year AI has to face reality.

Then, Musk continued to add that this was the third case of a CEO running away in just five months, and the CEO of Adept had just jumped to Amazon not long ago.

If this isn’t evidence that large, well-funded AI startups are struggling, then what is?

This is also an obvious signal: insiders seem to have quietly realized that LLM has reached a bottleneck. If AGI is really only one or two years away, who would switch jobs?

Marcus said that he had predicted everything that is happening now a year ago.

Finally, Marcus said with certainty: The AI ​​bubble may burst within a few weeks.

$1 trillion is being spent on AI, but is the demand enough?

Coincidentally, Business Insider recently wrote a report that a pharmaceutical company CIO chose to cancel Office 365 Copilot after letting employees try it for 6 months.

The reason is simple—he believes that these so-called AI capabilities are not worth paying for.

This anecdote gets to the heart of the generative AI craze — and whether it can continue.

It is predicted that technology giants will spend $1 trillion on data centers, real estate, chips and other supporting equipment to build AI models, tools and products.

But these investments will only pay off if there is real demand for products and services, especially from corporate customers.

The CIO’s story is clearly not a good sign — if a pharmaceutical company can’t spend an extra $180,000 on these AI tools, that’s a problem for the entire tech industry.

For more than a year, generative AI has been expected to unleash a huge wave of new demand, and companies and venture capitalists have been burning through cash reserves in pursuit of the trend.

However, what if this demand is weaker than expected?

The risk of overinvestment

The most obvious point is the construction of data centers, which is very expensive and mostly fixed.

After all, when you find that you don’t actually need so much infrastructure, you can’t make it “disappear on the spot”.

Amazon CEO Andy Jassy put it well during the company’s earnings call:

If you don't have enough capacity, then you might have service outages. Obviously, nobody wants to do that. But if you have too much capacity, the economics are terrible.

Google CEO Sundar Pichai said the company would choose to overinvest rather than miss out on potential AI revenue opportunities:

For us, the risk of underinvesting far outweighs the risk of overinvesting, even if we ultimately prove to be overinvested.

Nadella and his 365 Copilot

However, compared to the pessimistic voices, Microsoft CEO Satya Nadella may be the most optimistic of all.

He said the software giant is seeing strong demand signals from customers --

Microsoft 365 Copilot is our best Office 365 or M365 suite, and many customers come back to buy more seats.

Obviously, the story told by Nadella is completely opposite to that of the dissatisfied pharmaceutical company CIO in the BI report.

The question is, is AI a bubble?

Investment is accelerating

According to the latest statistics from The Information, the GenAI craze is far from over.

This year, GenAI startups raised a record $12.2 billion in the second quarter, surpassing the first quarter of 2023.

Lightspeed has been one of the most active firms, leading investments in at least five companies, from Skild, which develops robot control software, to Suno, which uses AI to generate songs.

Accel also led at least five rounds, including $1 billion in Scale AI, $220 million in model development company H, and $90 million in image generation company Synthesia.

The previous record was mainly due to Microsoft's $10 billion investment in OpenAI, while this time it was Musk's xAI that raised $6 billion in May.

But even excluding xAI, funding for GenAI startups increased by more than 85% compared to the same period last year.

At the same time, the number of companies receiving financing exceeded any quarter since The Information began recording in 2021 - a total of 55.

Among them, the companies with the most financing are basically training basic models.

For example, xAI, which was just mentioned, Mistral AI, an open source model development company that raised $640 million, and Cohere, which raised $500 million.

Similarly, startups that are committed to making model training and use easier are also a focus of venture capital.

For example, Scale AI, a company that labels training data, raised $1 billion in May.

In comparison, there are fewer areas of consumer AI applications.

Grace Isford, partner at Lux Capital, said, "AI applications are still in their early stages, while the infrastructure layer is relatively more mature. At the same time, companies that run large-scale models also need more capital."

Guru Chahal, partner at Lightspeed Venture Partners, also said, "Investors are more willing to invest heavily in companies that develop models and related software because the scale of the opportunities they are pursuing is also larger."

However, among these companies that are developing AI applications, more and more are beginning to develop their own models that are more suitable for their own business needs.

For example, Pika Labs, which uses AI to generate videos (raised $135 million in June), Suno, which uses AI to generate songs (raised $125 million in May), and Skild, which develops robot control software (raised $300 million in July).

Of course, these companies are not doing it all alone. Open source models from companies like Meta and Mistral are popular starting points for building their own models.

According to statistics, there are currently more than 100 companies building their own AI models.

The second most popular model is OpenAI, which is used by 68 startups.

In addition, several selected models from multiple suppliers.

Investors are also rethinking

But the question is when this technology can actually bring benefits to businesses.

Over the past two weeks, tech giants have reported to shareholders one by one on their huge investments in AI, which have led to a surge in capital expenditures.

According to S&P Global Market Intelligence, Microsoft and Amazon's capital expenditures in the second quarter increased by more than 50% compared with the same period last year; Google's increased by more than 90%; and Meta's increased by more than 30%.

Compared to these expenditures, the revenues from these investments remain relatively meager.

As of Q1 2024

AI startups also seem to be quickly realizing how difficult it is to generate enough revenue in the short term to pay the bills.

As mentioned above, this is the third time we’ve seen founders of AI startups abandon their startups for larger tech companies.

Character.AI, which was just "sold" to Google, gave investors a return that was not too bad, but it was far from the initial expectations.

If these “flying away” founders don’t see significant business challenges, it’s hard to understand why they would jump ship.

To be sure, some companies are starting to see returns on their AI spending.

ServiceNow, for example, last week raised its fiscal year revenue forecast due to significant progress in its AI products.

But given the brutal performance these companies have had over the past year, it could be a roller-coaster ride for investors.

References:

https://www.theinformation.com/articles/google-hires-character-ai-cofounders-and-licenses-its-models?rc=epv9gi

https://www.theinformation.com/articles/ai-investors-are-soul-searching

https://www.theinformation.com/articles/pro-weekly-investments-in-generative-ai-accelerate

https://www.businessinsider.com/genai-jitters-is-there-enough-demand-1-trillion-ai-spending-2024-8