Grandfather of AI describes Musk’s xAI as a “failure” and warns of a bubble

Yann LeCun aka “Godfather of AI” sharply criticizes Elon Musk’s AI company xAI and warns of a massive bubble in the entire industry. The enormous operating costs for language models are disproportionate to the revenue, says LeCun.

Criticism of xAI and the industry

Yann LeCun, AI pioneer and current chief AI scientist at Meta, considers the company xAI, founded by Elon Musk, to have failed, partly because the company has seen numerous departures from the team. At the same time, the scientist warns of difficult economic developments in the entire artificial intelligence industry, as operating costs currently significantly exceed revenues. The working atmosphere at xAI makes it difficult for Musk to attract qualified developers, says LeCun. To make matters worse, the company is recording high financial losses. In the first quarter of the year, xAI reported an operating loss of $2.5 billion (around €2.17 billion).

Danger of a market bubble

As the AI ​​researcher said in an interview with CNBC explains, the financial imbalance does not only affect Musk’s project. Providers like OpenAI or Anthropic are currently heavily subsidizing the use of their services with investor money. Although prices for end customers are rising, the expenses for operating data centers are not falling quickly enough.

If providers do not reduce their costs or increase prices for users, there is a risk of an investment bubble bursting. According to the expert, the previous large language models also have structural problems:

New approaches in development

As an alternative to language models, LeCun relies on world models. Such systems are designed to learn how the real world works, including physical laws and cause-effect principles. Research into the approach requires large investments. In some cases, sums of around a billion dollars (around 868 million euros) in fresh capital are flowing into projects that develop such new architectures. Large language models offer advantages in programming, text generation or mathematics. However, the high computational effort often does not justify the benefit for the user. It remains to be seen whether the World Models architectural approach will prove to be more practical in the long term.