“And in the end, OpenAI doesn’t matter. They are making the same mistakes as we are in their posture relative to open source, and their ability to maintain an edge is necessarily in question,” reads Google’s internal leaked document titled, “We Have No Moat, and Neither Does OpenAI.”
In the leaked document, it also said that it is definitely worth noting that ever since developers got their hands on Meta’s leaked LLM, LLaMA, the open-source community is flooded with LLMs-based generative AI models. Looks like Meta’s “mistake” has actually brought them back into the race, but only through open-source.
Interestingly, Google claims that OpenAI has no moat when it comes to LLMs. But among the hype around ChatGPT, we are increasingly seeing startups that are either using GPT in their name, or building technologies using the APIs provided by OpenAI.
During all this, Google backed a lot of AI startups to fight against OpenAI and Microsoft. It has been trying to push generative AI into its cloud infrastructure, and thus partnered with Salesforce, Box Inc, Jasper AI, and Canva. The companies have been using Google’s LLM offering through Vertex AI on Google Cloud.
OpenAI Has A Moat
Interestingly, despite the partnership with Google, last month Salesforce also announced Einstein GPT, integrating OpenAI’s technology and offering generative AI capabilities for its customers.
In the same announcement, Salesforce also announced a $250 million generative AI fund which will first invest in four companies that include Cohere and Anthropic, both of which are backed by Google; and the companies will use GPT technology on Salesforce’s platform. Seems like the startups backed by Google are using its funds on OpenAI’s technology to run their businesses.
Another Indian startup, Slang Labs, which is a voice assistant and search based, is backed by Google. Interestingly, the startup announced the launch of CONVA 2.0, a multilingual AI co-pilot which is powered by GPT technology, leveraging it for e-commerce shoppers. The company also boasts that their technology with GPT offers 46% better performance than Google’s current Voice Assistant.
In an interview with AIM, Slang Labs’ CEO and co-founder, Kumar Rangarajan said that the company decided to use GPT into its products to offer services to its customers regardless of who developed it. “Making LLMs from ground is a very expensive and computation heavy process. It does not make sense to do it, therefore we adopted the GPT technology into our services,” Rangarajan explained.
Similarly, Anthropic, an AI startup backed by Google also partnered with Notion, which is a conversation chatbot actually a wrapper of OpenAI’s GPT-3.5. The businesses just want the best services for their customers, and it seems as though OpenAI is providing them with that.
Opensource: Google Killer?
This should be concerning for Google that some of its startups are also leveraging its rivals’ technologies. Looks like GPT is the moat for OpenAI. And where OpenAI is lagging behind with a lot of its open-source policies, privacy issues, and many more concerns, Meta’s LLaMA is filling the gap with its ability to allow individuals to build local models.
“The modern internet runs on open source for a reason,” reads the document. “And we should not expect to be able to catch up [to open source].”
Google likened the recent open-source-based AI development to the Internet. It acknowledged that in the Internet technology revolution, the open-source community has a bigger role to play and cannot be replicated by any company. “Individuals are not constrained by licenses to the same degree as corporations,” read the leaked document from Google.
The only thing that the big-tech has is computation capabilities, which are also not required since the introduction of LLaMA based models that can build ChatGPT like models on a single computer. Probably, instead of fearing OpenAI and GPT, Google is more worried about open-source.
“But the uncomfortable truth is, we aren’t positioned to win this arms race and neither is OpenAI. While we’ve been squabbling, a third faction has been quietly eating our lunch — open-source.” But is it true that OpenAI and Microsoft are on the same level as Google, without a moat? It might not be true, Google.
Can Google Catch Up?
A lot of the development has happened, probably, because most of the startups have already started building on top of GPT. There was no other offering other than GPT before. “Every developer I know is building on top of GPT,” said Robert Scoble in his tweet talking about what Google along with Apple should do to get ahead in the game.
For example, Khan Academy, one of the first companies’ to adopt GPT-4 into its offerings built his entire system on GPT-4. Would there be any reason for him to make the change and adopt Google or Apple’s offering in the future, unless they are a hundred steps ahead of OpenAI and Microsoft’s?
Surprisingly, Google I/O 2023 was a hit. Sundar Pichai and team introduced several advancements in their AI systems, including generative AI and LLMs. It also announced the launch of the PaLM-2 language model along with its API. It also hinted at the multimodal Gemini project that the company has been working on with DeepMind.
On the flip side, the people who have already been building on top of GPT products would never want to shift to what Google is going to offer in the near future, such as PaLM-2, unless it is worth it. Same is the case with Apple if it gets into the LLMs field. We will hopefully hear more about this at Apple’s conference in June.
For now, Meta’s LLaMA has won a lot of developers on its side, but apart from that, the case is still that as soon as someone hears generative AI or chatbot, they hear GPT or OpenAI. Google needs to step up its game, and the Google I/O conference was definitely one step in the right direction. Anyway, OpenAI’s GPT is built using Google developed Transformers.