Open AI vs. Google – No lunch

OpenAI and Microsoft will be unable to eat Google’s lunch

• Having failed to move the search market with ChatGPT, OpenAI is developing a product specifically designed to replace search, but I suspect that GPT’s inability to understand causality will mean that it fails to make a dent.
• Should Google’s shares undergo a big correction as a result of the endless hype and chatter, this would be a good entry point.
• OpenAI appears to be developing a search algorithm that would power Bing and bring its performance into line with Google giving it the opportunity to break the biggest internet monopoly.
• However, it is one thing to plan a product but quite another to make it work as projected.
• This is where I think that OpenAI is going to fail because the last 18 months have demonstrated that LLMs are really bad at search because they have no idea of what it is that they are doing.
• I have compared Google search against a number of GPT-based systems connected to the internet and Google destroys them every time.
• Google has been doing search for 25 years and in that time, it has evolved an uncanny ability to work out what the user is searching for even if the user inputs the terms unclearly or incorrectly.
• In my opinion, this is a demonstration of causal understanding of what the user is searching which is what allows Google to perform the search so well and also to dig up obscure but relevant items (the long tail).
• Google does not have one massive neural network that performs search but a large range of algorithms which are stitched together with rules-based software.
• Rules-based software differs from neural networks in that while it cannot learn, it can reason which allows the Google algorithm to inject some causality into the search that it is performing.
• OpenAI’s approach has long been that with more data and more compute, the answer will magically pop out at the end which is why over the years its models and its consumption of capital have got bigger and bigger.
• However, while its machines are very good at conversing using natural language, they have no causal understanding of what it is that they are doing, and it is here where I think that OpenAI’s ambitions for search will come unstuck.
• OpenAI can fine-tune GPT-4 as much as it wants to enhance its capability for search, but the nature of search is that one has to incorporate new information into the data set in real-time.
• This is something that all systems based on deep learning are really bad at and so I don’t think that an LLM-based approach is going to realistically be able to solve search nearly as well as Google can.
• Furthermore, Google also has LLMs that are just as capable as OpenAI’s and so if this was the answer to search, then Google would have made use of it.
• Hence, when OpenAI and Microsoft launch their new search product to much fanfare and acclaim, there is a good chance that Google’s valuation will take a hit.
• Google is already by far the cheapest of all of the big technology stocks with a 2024 PER of 21.2x and 2025 of 18.4x and it is still the indisputable global leader in AI in my opinion.
• This is already a reasonable valuation and so if it was to get cheaper as a result of market-driven panic about how OpenAI and Microsoft are about to eat Google’s lunch, then this would represent a good opportunity.
• I don’t own Google at the moment but am tempted to have a proper look at it in anticipation of an AI-driven panic.