Generative AI took the world by storm in November 2022, with the launch of OpenAI’s ChatGPT service. One hundred million people started using it, practically overnight. Sam Altman, CEO of OpenAI, the company that created ChatGPT, became a household name. And at least half a dozen companies competed with OpenAI in an effort to build a better system. OpenAI itself sought to surpass GPT-4, its flagship model, presented in March 2023, with a successor, presumably called GPT-5. Virtually every company scrambled to find ways to adopt ChatGPT (or similar technology, made by other companies) into their businesses.
There’s just one thing: Generative AI doesn’t actually work that well, and maybe never will.
Fundamentally, the engine of generative AI is fill-in-the-blank, or what I like to call “autocomplete on steroids.” These systems are great at predicting what might seem good or plausible in a given context, but not at understanding at a deeper level what they are saying; an AI is constitutionally incapable of verifying its own work. This has led to massive problems with “hallucinations,” in which the system states, unreservedly, things that are not true, while inserting stupid errors into everything from arithmetic to science. As they say in the military: “often wrong, never in doubt.”
Systems that are often incorrect and never questioned make fabulous demos, but are often terrible products in and of themselves. If 2023 was the year of AI hype, 2024 has been the year of AI disillusionment. Something I argued in August 2023, in the face of initial skepticism, has been felt more frequently: Generative AI could be a failure. The profits are not there.estimates suggest that OpenAI’s operating loss in 2024 may be $5 billion, and the valuation of over $80 billion does not align with the lack of profits. Meanwhile, many customers seem disappointed with what they can actually do with ChatGPT, compared to the extraordinarily high initial expectations that had become commonplace.
Also, basically all the big companies seem to be working from the same recipe, creating bigger and bigger language models, but all ending up in more or less the same place, which are models that are as good as GPT-4, but not a far better. What that means is that no individual company has a “moat” (a company’s ability to defend its product over time), and what that in turn means is that profits are declining. OpenAI has already been forced to reduce prices; Now Meta is giving away similar technology.
As I write this, OpenAI has been demoing new products, but hasn’t actually released them. Unless it comes up with some major breakthrough worthy of the GPT-5 name before the end of 2025 that is decisively better than what its competitors can offer, it won’t be all that come. The enthusiasm that underpinned OpenAI will wane, and since it is the model for the entire field, it may all soon fail.