Make Up Your Mind

Yesterday I wrote about Google and Microsoft trying to compete on each other’s turf by adding generative AI to their search and productivity products.

CNN reported later yesterday that Meta is also planning to go all-in on AI, adding generative AI to their products (and apparently pivoting from their focus on the metaverse). I called it a “war” yesterday, as in a turf war over search and productivity tools, but the bigger picture of what’s happening this week is a race by incumbents, start-ups, investors, and everyone else, to embrace generative AI.

https://www.cnn.com/2023/03/15/tech/meta-ai-investment-priority/index.html?utm_content=2023-03-15T14%3A55%3A02&utm_term=link&utm_medium=social&utm_source=twCNN

In contrast to the “two companies, two fronts” I wrote about yesterday, the bigger picture taking shape this week is that the starting gun has been fired, and multiple races are taking place at multiple levels.

1. AGI Race – The race toward Artificial General Intelligence with the release of ChatGPT4
2. Start-up Race – The race to start new AI companies, to take advantage of the new technology, for example, to capture “vertical markets” – tailoring LLMs to specific business areas, such as law, pharma, and learning tools.
3. The Platform Race – The race to establish a generative AI “model” that can become a platform for companies pursuing vertical markets.

The AGI race is the least well-defined of the three big areas, in that no one is in sight of achieving AGI, and not many companies are claiming it as their goal. But OpenAI does clearly state that their goal is to create AGI, and they point to ChatGPT4 as the leading technology heading in the direction of AGI.

There is no consensus on the exact definition of AGI, but OpenAI’s Sam Altman describes it as the ability to do any useful task that a human of average intelligence can do. GPT4 has completed a battery of tests such as the SAT and the LSAT, scoring above the average of humans taking those exams. But this is “book smarts” on steroids, in that the AI doesn’t actually understand the concepts. It is just very good at using its training on massive amounts of information to predict answers to questions (posed in text, a form of test the LLM is built for).

Gary Marcus (NYU Professor Emeritus) is a cognitive scientist who points out the limitations of the current generation of LLMs, and argues that they are not structurally capable of being scaled up to AGI. The release of GPT4 has energized an ongoing debate on this front. I’ll write more on this later this month.

The Start-up Race is well-defined on two fronts. First, the race to find top AI scientists who are interested in founding a new company, or going to work for a start-up. The reason this is well-defined is that there is a limited pool of AI experts who are in the top tier of the field. And AI is the sort of domain where the difference between top-tier and average is huge.

The other aspect that is fairly well-defined is the race for vertical markets. OpenAI plans to provide an Application Programming Interface (API), which will enable start-ups to build domain-specific tools, such as an AI legal assistant. The vertical market company will focus on the legal field, tailor the front end of the tool for lawyers, but will not have to build a language model of their own. They will connect to OpenAI’s ChatGPT4.

The Platform race is also somewhat well-defined, in that there are only a few large companies who can compete with OpenAI, which has deep pockets via its partnership with Microsoft, who recently pledged 10 billion dollars to OpenAI. Training a LLM on the scale of ChatGPT4 is very expensive, especially in computing power. It takes the largest (most expensive) supercomputers to train the models on the scale of GPT and its competitors.

Google and Meta have LLMs of their own, and Google especially, is likely to take on OpenAI for market share in the AI platform race. The infrastructure that Google and Microsoft have in place for cloud computing puts them in a stronger position than Meta. Amazon announced a partnership with start-up Hugging Face, which is competing with OpenAI in the generative AI space. Since Amazon is a dominant player in cloud computing, and use AI extensively, they are clearly in position to compete in the platform race. Stay tuned, the race is on!