Generative AI took the world by storm in November 2022, with the release of OpenAI’s ChatGPT service. One hundred million people started using it, almost overnight. Sam Altman, CEO of OpenAI, the company that developed ChatGPT, became a household name. And at least half a dozen companies ran on OpenAI in an effort to build a better system. OpenAI itself wanted to replace GPT-4, its flagship model, which was launched in March 2023, with a successor, likely to be called GPT-5. Almost every company struggled to find ways to use ChatGPT (or similar technology, developed by other companies) in their business.
There’s just one thing: Generative AI doesn’t actually work that well, and probably never will.
Basically, a generative AI engine fills in the blanks, or what I like to call “autocomplete on steroids.” Such programs are very good at predicting what may sound good or sound in a particular situation, but not at a deep level of understanding what they are saying; AI is constitutionally incapable of fact-checking its own work. This has led to serious problems of “illusion,” where the system asserts, without merit, things that are not true, while making heady mistakes in everything from mathematics to science. As they say in the military: “often wrong, never doubt.”
Systems that are always wrong and no doubt make great demos, but are often crappy products themselves. If 2023 was the year of AI hype, 2024 was the year of AI disappointment. What I argued in August 2023, in the initial skepticism, has been heard over and over again: the production AI may turn out to be a dud. The profit is not there—estimates suggest that OpenAI’s 2024 operating loss could be $5 billion—and the calculation of more than $80 billion does not match the lack of profit. At the time, many customers seemed disappointed with what they could actually do with ChatGPT, related to the unusually high expectations that had become the norm.
In addition, basically all the big companies seem to be working in the same way, making bigger and bigger versions of languages, but they all end up in the same place, which are models that are almost equal to GPT-4, but not much better. What that means is that not every company has a “moat” (the ability of a business to protect its product over time), and what that means is that profits are diminishing. OpenAI has already been forced to cut prices; now Meta offers the same technology for free.
As I write this, OpenAI has been dropping new products but not actually releasing them. Unless it comes out with many improvements worthy of the GPT-5 name before the end of 2025 that are better than what its competitors can offer, the bloom will be off the rose. The enthusiasm behind OpenAI will wane, and since it’s the poster child for the entire industry, everything could go fast.
Source link