According To Keras Creator François Chollet, Investments In AI Are "1000 Times Too High"

The Swiss newspaper Neue Zürcher Zeitung, popular among conservatives in the German-speaking world, published a remarkable interview with AI researcher and Keras creator François Chollet on March 4.

In it, Chollet makes it clear from the start that he thinks the current hype about the new Large Language Models is "exaggerated". He initially justifies this by saying that people in San Francisco are talking about realizing artificial general intelligence in two years, but he prefers to leave the context open. Instead, he goes on to say that AI will not reduce the need for programmers, despite the current massive layoffs in the tech industry. On the contrary: Chollet even claims that "more programmers will be needed in 5 years than today". Once again, he fails to provide the reader with any further explanation.

This is followed by a very interesting economic analysis: according to Chollet, earnings from AI will amount to $6 billion in 2024, while $90 billion has been invested in AI in the last two years. In the next paragraph, he concludes that the investments in AI are "by a factor of 1000" too high and completely "decoupled" from reality. Unfortunately, he does not tell the reader what original valuation method he uses to arrive at the eye-popping factor of 1000 based on the numbers he cites.

Perhaps this is a good time to ask one of the AIs that Chollet claims are not really intelligent, such as ChatGPT (I used GPT 4 for this): According to Chollet, AI earnings were $4 billion in 2023, will be $6 billion in 2024, and could be $10 billion in 2027 (Chollet does not cite sources for this). I asked ChatGPT to value the AI industry based on these numbers. As is common in asset valuation, ChatGPT used a discounted free cash flow model. Since Chollet's numbers already imply flat growth from 2023 to 2027, ChatGPT estimated growth from 2027 to be a sad 5%. This led ChatGPT to an extremely conservative valuation of the AI industry of $156.3 billion. So the alleged $90 billion would be a phenomenal investment, wouldn't it? Neither artificial intelligence nor my own limited natural intelligence comes up with a thousandfold overvaluation here, what a shame!

But enough of splitting hairs, Mr. Chollet has other things to tell us. For example, that generative AI is only a matter of "large databases" and not intelligence, since it cannot perform transfer tasks, but only reproduces memorized information. I have given ChatGPT the most absurd programming tasks and have been surprised by the results every time. I doubt very much that source code exists for the hair-raising programming tasks I have given ChatGPT over the past year.

Chollet's answer to the question of whether society can adapt to the current technological upheaval is also exciting - according to Chollet, it doesn't have to, because "the changes between 1870 and 1920, for example, were much greater". Much greater than what, Chollet leaves open. Much greater than the changes from 1970 to 2020? Much greater than the changes since the release of ChatGPT, i.e. since November 2022? As before, Chollet makes some claims and doesn't even bother to justify them. In this interview we learn more about Mr. Chollet than about the technology he considers overrated. It seems that the once-celebrated wunderkind has not quite digested the fact that he is no longer at the forefront of the current AI revolution.