Expert warns AI improvements have plateaued, financial bubble pop incoming

TL;DR: AI companies like OpenAI and Microsoft are racing to develop artificial general intelligence (AGI), but experts like Gary Marcus suggest it may not be economically viable. Marcus argues that the AI market is a bubble, with valuations based on the assumption that scaling Large Language Models (LLMs) will lead to AGI.

Artificial intelligence companies such as OpenAI, Microsoft, and other big players such as Elon Musk’s X-based AI Grok are all in a race to develop artificial general intelligence (AGI). But what if no one makes it to the finish line because it’s not economically viable?

Expert warns AI improvements have plateaued, financial bubble pop incoming 363636

VIEW GALLERY – 2 IMAGES

This suggestion has been made by several experts and is a theory held by Gary Marcus, a cognitive scientist and AI skeptic. Marcus says the current AI market is a bubble and that people haven’t yet realized the valuations of companies such as OpenAI and Microsoft are predicated on the bet that the underlying technology powering AI systems, Large Language Models (LLMs), will grow in power with scale. However, that might not be the case.

The Information reported last week that OpenAI researchers discovered the company’s latest AI model, codenamed Orion, was a noticeably less improvement in performance compared to the leaps made from GPT-3 to GPT-4. IIya Sutskever, co-founder and former chief science officer of OpenAI, previously told Reuters that improvements from scaling AI models bigger have plateaued, which harms the general theory that the bigger the AI model the more powerful.

The economics are likely to be grim,” Marcus wrote on his Substack. “Sky high valuation of companies like OpenAI and Microsoft are largely based on the notion that LLMs will, with continued scaling, become artificial general intelligence.

With this in mind, Marcus suggests that given the cost of the AI chips to run these models, the power it takes to fuel the data centers, and the cost of training, the AI market will eventually reach a point where it won’t be economically viable to invest more money into training more complex AI models.

The economics will likely never make sense: additional training is expensive, the more scaling, the more costly,” said Marcus

LLMs such as they are, will become a commodity; price wars will keep revenue low. Given the cost of chips, profits will be elusive. When everyone realizes this, the financial bubble may burst quickly,” added Marcus