Generative AI took the world by storm in November 2022 with the release of OpenAI ChatGPT. One hundred million people started using it practically overnight. Sam Altman, CEO of OpenAI, the company that created ChatGPT, has become a household name. At least six companies have raced OpenAI to try to build a better system. OpenAI itself has been trying to outdo GPT-4, its flagship, introduced in March 2023, with a successor likely to be called GPT-5. Virtually every company has tried to find ways to implement ChatGPT (or similar technology developed by other companies) into their operations.
There’s just one thing: Generative AI doesn’t really work that well, and perhaps never will.
Essentially, a generative AI engine is fill-in-the-blank, or what I like to call “autofill on steroids.” Such systems are great at predicting what might sound good or credible in a given context, but they don’t understand at a deeper level what they’re saying; artificial intelligence is constitutionally incapable of checking its own work. This has led to massive “hallucination” problems, in which the system states unqualifiedly things that are not true while introducing blatant errors in everything from arithmetic to science. As they say in the army: “wrong often, never doubt.”
Systems that are often wrong and about which there is no doubt make great demos, but are often impoverished products themselves. If 2023 was the year of AI hype, then 2024 was the year of AI disillusionment. Something I argued in August 2023, to initial skepticism, is being felt more often than not: Generative AI may be a dud. There are no profits there –estimates suggest that OpenAI’s operating loss in 2024 could be $5 billion, and a valuation of over $80 billion does not reflect the lack of profits. Meanwhile, many customers seem disappointed with what they can actually do with ChatGPT, compared to the extremely high initial expectations that have become common.
Moreover, it seems that basically every enormous company is working to the same recipe, creating larger and larger language models, but it all ends up in more or less the same place, which is models that are about as good as GPT-4, but not much better . This means that no single company has a “moat” (the ability of a company to defend its product over time), and this in turn means that profits decline. OpenAI has already been forced to cut prices; now Meta is giving away similar technology for free.
As I write this, OpenAI is testing fresh products but not actually releasing them. Unless there is some significant advancement worthy of the GPT-5 name before the end of 2025, it will flourish by far being better than what its competitors can offer. The enthusiasm that supported OpenAI will wane, and since it is the industry’s flagship model, the whole thing could soon collapse.