OpenAI is getting stronger and stronger, not only stealing business from tech giants like Google, but also smashing the jobs of entrepreneurs.
After the first developer conference, many developers worried whether their projects would be replaced. And before the meeting, several AI companies had already started layoffs, including many star projects that had taken huge amounts of financing.
Meanwhile, OpenAI is accelerating its moneymaking. It launched a paid version of ChatGPT Plus, went online with the GPT Store, and turned into an increasingly utilitarian commercial company. Last year OpenAI’s revenue was in the tens of millions of dollars; this year it will exceed $1 billion.
All of this seems to be drifting away from its former mission as a non-profit organisation with general artificial intelligence as its ultimate goal. the sudden removal of CEO Sam Altman reflects a disagreement among OpenAI’s management over the company’s philosophy of growth. The empire has not yet been built, and internal cracks are already appearing.
For entrepreneurs, how to deal with the relationship between them and OpenAI will be a critical and delicate issue.
One AI entrepreneur said: either do what OpenAI doesn’t do, or do what OpenAI can’t do yet. Some people are beginning to actively change their strategies to find a suitable space for their own survival.
After the short window, the space left for some entrepreneurs seems to be running out.
“Shell” business, can not be done for a long time
More and more GPT shell projects are being replaced by GPT.
The word “shell” is derogatory in the domestic AI circle. When many AI companies hear “shell”, they wave their hands for fear of getting involved. However, in foreign countries, the shell is very normal, and related start-up projects are endless.
Since the birth of the GPT model, there has been a contradiction in the industry: there is a huge gap between the public’s growing demand for large models and the complex technology and difficult operation.
The companies that bridged this gap grabbed the market lead and grew extremely rapidly for a while.
Typical companies such as Jasper, a marketing copy generation company, was founded in January 2021, valued at $1.5 billion in 18 months, and surpassed OpenAI in revenue last year.
Jasper’s model is very simple: call the ability of the GPT-3 model, fine-tune the model with marketing data, create a user interface as the front-end of the model, and the user interacts with the model through this interface to generate various styles of marketing copy. Users didn’t need to know what GPT-3 was, let alone understand the complex technology behind it.
The problem with GPT-3 at that time was that it couldn’t talk to the user directly, and it needed very precise prompt words to get excellent answers. jasper productised the GPT-3 model and made a very good on-boarding front-end, which was equivalent to putting a shell on GPT.
There are many similar products. The one that changes essays, the one that writes code, the chatbot, the virtual assistant …… A lot of xxx GPTs that popped up in the first half of this year are all shell products.
Products that are simply shells are easily replaced.
The first wave of impact came from ChatGPT. launched at the end of November last year, ChatGPT, text processing capabilities are very stunning, two months harvest 200 million monthly users. jasper and other companies have alternatives to their products.
Then ChatGPT function gradually improved. the end of October, ChatGPT began to be able to directly read PDF files, automatically browse the web, data analysis, ChatPDF, AskYourPDF, PDF.ai and other companies threatened by the business.
An AI entrepreneur said to “fixed focus”, this feeling, like a large shopping mall, the operator in the business of the best shops on the side of the shop, opened a similar shop, in all aspects of more support.
The second wave of impact comes from GPT Builder. at the developer conference in November, OpenAI launched GPTs, which allow people who don’t know code at all to customise their own GPT apps, which can do web searches, make charts, do data analysis, and complete a variety of tasks.
This is equivalent to, ordinary people can also “shell” to create GPT.
GPT has become ubiquitous. Copywriting, knowledge quizzes, virtual assistants, code collaboration, text-to-diagrams and other verticals have been ruthlessly crushed.
If a company is just doing very shallow interface design and feature integration on top of the GPT model, it risks being replaced. jasper has already had a wave of layoffs in July, and there are many more like it.
For those products that don’t have a shell, but overlap with ChatGPT’s capabilities, they also face serious challenges.
In the past six months, Grammarly, an AI writing assistant company, Deepgram, an AI automatic speech recognition company, and Stack Overflow, a Q&A platform for programmers, have all made layoffs. And their products existed years before ChatGPT.
These companies’ products were valuable early on, but when OpenAI launched similar products, their functionality and value were quickly diluted.
“Tool-based products, are the easiest to replace, especially for companies like OpenAI that have mastered the underlying technology, copying a pop-up app is a matter of minutes.” An AI entrepreneur said to “fixed focus”.
Eliminate you, it has nothing to do with you. This is the inevitable evolution of OpenAI.
The space for middlemen is getting smaller and smaller.
There are middlemen in any industry, and AI is no exception.
From the technical architecture, the AI model can be roughly divided into three layers: model layer, platform layer, application layer.
The platform layer, also called the tool layer, refers to a series of engineering capabilities required in the middle to integrate the big model into the application, playing a bridging role. For example, based on GPT-4 as the underlying big model, to develop ChatGPT-like applications, we need to use various tools in the platform layer.
There are many tool stacks in the platform layer, such as the development tool chain (Langchain) and the model tool chain (for data annotation, vector database, distributed training, etc.). A large number of startups have emerged on the chain, with companies doing vector databases seeing intensive financing this year, and companies doing toolchains represented by LangChain are hot.
The significance of the existence of these companies lies in the fact that there are still a lot of imperfections in the big language model, the development system of the whole ecosystem has not yet been formed, and when developers deploy the language model for production, it is not enough to simply rely on the cue words, and they need to be supported by more underlying tools, and the vendors of the big language model, such as OpenAI, have failed to provide these tools in a timely manner.
LangChain, launched at the end of October last year, is an open-source Python and JavaScript library that encapsulates a large amount of application development logic and tool integration for large language models. In short, it can answer questions with reference to an entire database, such as accessing the latest data, reports, documents, and website information, connecting various sources of information.
As we know, when ChatGPT first launched, it was not searchable online, and the data was only updated to 2021; LangChain bridges the gap between it and external data.
LangChain itself does not develop big models, but helps developers use big models. It is like an important intermediary station, integrating a variety of commonly used tools and components, making it easy to develop applications. Since its launch, it has been highly sought after by developers and has become an important LLM application development framework.
There’s no doubt that LangChain solves developers’ problems. But when OpenAI decided to go a step further and what was once a problem was no longer a problem, the space for middle tier entrepreneurs was squeezed.
OpenAI’s newly released Assistant API is an LLM-based development framework for developers. Through it, developers can call all functions including data analysis, function call, image recognition, speech generation, etc.. The difficulty of development has been greatly reduced. “One-stop development” started to become a reality.
This makes LangChain and other middle-tier companies in an awkward position.
Atom Capital, an investment organisation, wrote: “A large number of Agent framework companies will lose their value of existence, and developers will move to OpenAI’s official framework for reasons such as ecological convenience.”
In the first half of the year, OpenAI personally ignited the enthusiasm of many tooling layer startups, and now it has personally doused it.
The Assistant API also has the ability to directly retrieve external data, automatically optimise it, and convert the developer’s own data into GPT’s knowledge base. In other words, GPT can do its own vectorisation of data, and those companies that make vector databases will have to rethink how much room they have in their business model. This is just the beginning.
The launch of the GPT Store shows that OpenAI is already building its own application ecosystem. It makes it easier for developers to develop their own applications, while allowing people who don’t know how to code to create AI agents based on their own knowledge base through natural language.
Sheng Wang, managing partner of InnoAngel Fund, analysed to “Focus” that this has led to the development of those companies based on LangChain’s framework in the past, which has become less competitive, and that OpenAI will hit a number of companies that develop Agents, such as a variety of counselling consultants, psychological counselling, knowledge explanation, etc. “Now these products are actually all very competitive. “Now these products are actually facing a great challenge, because OpenAI has already done all these (technical framework), you just focus on doing a good job on the content.”
What OpenAI won’t do
Because of ChatGPT, OpenAI has become a product company for the average user, competing with the companies that do business on its platform. But at the same time, it remains a platform for developers and tries to be the birthplace of AI-native applications.
Many people only know about the chatbot ChatGPT, but in fact OpenAI has also launched three products - DALL-E, a text-to-image tool, Codex, a natural language-to-code system, and Whisper, an automated speech recognition system. Taking the Codex model as an example, Microsoft launched AI in 2021 based on it. Take the Codex model as an example, Microsoft launched Copilot, an AI auto-programming tool, in 2021, killing a group of startups making AI programming tools on the beach.
Today ChatGPT is getting stronger and is replicating that story. So is there anything OpenAI won’t do, or doesn’t want to do for now?
The first is a companion agent, OpenAI has clearly said that it will not work in this direction, former CEO Sam Altman believes that human-like agents are not valuable, the real value is to assist people to complete their work.
That’s why ChatGPT is used by many people to code and write papers, rather than as a “friend” to chat.
AI companionship is a huge market. Ai’s impressive user growth and rising valuation of celebrity chatting app Character.Ai has fully verified the market space, and AI virtual chatting social software Glow has achieved a record of nearly 5 million users in four months since its launch.
If OpenAI doesn’t do it, there will be other companies competing for this market, whether they are based on their own big models or GPT models.
Logenic AI co-founder Li Bojie believes that “to do companion bot need to have core competitiveness, must not rely only on cue words, at least have their own fine-tuning model, have their own pipeline (pipeline), as well as infra (infrastructure) that can reduce the cost of reasoning.”
Also, games are out of OpenAI’s sights for now. Many believe that games will be deeply integrated with AI macromodels, that AI-driven NPCs (gaming terminology: non-player characters) will be given a digital life, and that the gameplay of the gaming industry will be transformed.
“If users can interact with game characters in natural language, and the plot is customised to the user’s liking, it will be a completely new gaming experience.” Li Bojie said.
MiniMax, the developer of Glow, got an investment from game company MihaYou, which is seen as MihaYou preparing for the upcoming changes in the gaming industry. startups such as MiniMax, which have the ability to develop large models and at the same time have a deep knowledge of the application scenarios, will have a unique advantage.
According to Li Bojie, there is also a category of things that OpenAI can’t do yet. For example, video input and video generation, Rewind’s recording pendant, similar to the film “Her” in the jacket pocket AI Pin such hardware-based products, relying on the smartphone Siri and so on is OpenAI is difficult to replace the entrance, there are data barriers to the scene is also very difficult for OpenAI to directly replace.
Data, as one of the three elements of the AI model, will become a high ground for vendors to fight for. openAI can’t collect data from all segments, and companies with data can also have a place.
In addition, to some extent, OpenAI lowers the threshold for ordinary people to participate in entrepreneurship in the field of AI. the emergence of GPTs directly creates a new profession - developers who don’t know how to code. People don’t need to write code, they just need to have ideas, insights and understand the market to create their own products.
“The GPT Store is most likely geared not towards developers, but creators, which is a deeply empowering, disintermediated strategy, and I think that’s the trend.” Someone commented.
Sheng Wang believes that today’s OpenAI is like Apple back then, representing a trend.After OpenAI’s developer conference, the industry has shifted from roll large models, to roll large model applications. “This signifies that the cycle of rolling big models may have passed, and next everyone has to innovate to develop applications.”
Starting a business at the application layer, OpenAI can’t do all areas. Just like Apple, music, video and other important applications, as well as some gadgets do it themselves, and leave the rest to the ecology.”
Taken together, OpenAI eliminates some opportunities while creating new needs. Those application layer companies that are highly dependent on API companies, and platform layer companies that have no competitive barriers, will face greater challenges in the future.
The only way to deal with this is to keep iterating and always adapt to the rapidly changing AI wave.
