Generative AI: Hype or Transformation?

　　Since the release of the groundbreaking generative AI tool ChatGPT in November 2022, investor interest in generative AI technology has surged. Technology stocks significantly outperformed the market, driving a sharp rise in the stock market in the first half of 2023. But is this technology truly transformative or has it been overhyped? Is it worthy of current investor enthusiasm?
　　Speaking of this, Gary Marcus, a leading figure in the field of artificial intelligence, quoted a Silicon Valley joke: A few years ago, if your startup had .ai in its domain name, you could calculate your valuation Add a zero. And now it feels like maybe two zeros, especially if you claim you’re using generative AI.
　　Does it sound like a bubble? “With the exception of the rapidly sinking and floating room-temperature superconductor LK-99, there are very few things I’ve seen that are more hyped than generative AI,” Marcus said. Many companies are valued in the billions. U.S. dollars, the entire market size is estimated to be in the trillions of U.S. dollars. Marcus poured cold water on this: the current actual revenue in a market that is said to be trillions of dollars is only a few hundred million dollars. These revenues may indeed grow more than 1,000 times, but this is just speculation. In an article titled “What if generative AI proves to be a dud? “In the blog post, he even asserted: The income from artificial intelligence has not yet arrived, and may never arrive.
　　Marcus said most of the revenue so far seems to come from two sources: writing semi-automated code (programmers like to use generation tools as assistants) and writing content. The former seems to be sustainable, but it’s hard to say in terms of content. Aside from chat-based search, which doesn’t work very well, there’s no killer app yet. Search is indeed attractive, but the technical problems are huge.
　　The underlying technology of generative AI (often called the base model because the system is fine-tuned on top of a large pre-trained model) is very unstable, making it difficult to design into a reliable product. You can never really predict whether a large language model will give you the right answer for any given question, and we now know that the answer can even change from month to month – which makes it difficult for third parties to integrate large language models into complex systems. Engineering became a daunting task. In some mission-critical cases, the engineering challenges may be insurmountable. Potential paying customers may quickly lose confidence as a result.
　　If Marcus is right, the entire field of generative AI may soon be over its current valuation. So far, valuations appear to be based on hopes and dreams without really taking into account serious engineering risks. Neither coding nor efficient but mediocre copywriting is nearly enough to sustain an inflated valuation. This is why after its initial popularity, the traffic of the ChatGPT website continued to decline. Global traffic to chat.openai.com fell by 9.7% and 9.6% in June and July of this year, respectively.
　　So, Marcus’ conclusion: “Even OpenAI will have a hard time achieving its $29 billion valuation; if it can only achieve tens or hundreds of millions of dollars in revenue year after year, that’s a valuation in the billions Competing startups below the dollar are likely to eventually fail. Microsoft, which has seen its stock price rise by nearly half this year, perhaps largely on its promise of generative AI, could plummet; NVIDIA has soared even more, but it could also fall back. ”
　　It’s almost to say that rapidly scaling technologies don’t always deliver on their promises, and if they don’t, the bubble can easily burst. Bubbles are often related to the value of eyeballs or clicks or the total addressable market, and these dynamics are talked about and create a sharing frenzy as a driver of valuation. Because the stakes are so high, investors never know if they’re in a bubble. However, this wave of artificial intelligence is still very different from the previous bubbles we have experienced.
　　Generally speaking, large technology cycles are initiated by upstarts. For example, as early as the early 1990s, when distributed computing was born, a small company called Oracle opposed the mainframe advocated by IBM. It took a while for this distributed technology to become more common because the giant IBM decided it represented the established way of doing things. Generally speaking, when switching from an old cycle to a new cycle, you will encounter established business practices and established technologies, which will hinder the adoption of new ways of doing things.
　　What’s different about this AI cycle is that it’s no longer led by an upstart. It’s driven by some of the most powerful tech companies on the planet. When technology providers unanimously agree that something is actually happening, it’s real, and when customers become interested, it’s not hype. Artificial intelligence’s vast potential would be possible if it were not caught in a hype cycle.
　　But of course, just because something is powered by AI doesn’t mean it’s going to make a ton of money. When it comes to generative AI, much of the current enthusiasm is driven by enthusiasm; bubbles create bubbles until they don’t. If some people get off the wagon and valuations start to drop slightly, it could set off a powerful negative feedback loop, causing values ​​that had been accelerating so rapidly to suddenly decelerate. Many talents and investors may flee generative AI for a host of new shiny things, just as some fled cryptocurrencies not too long ago.