August 3rd, 2024

Why the collapse of the Generative AI bubble may be imminent

The Generative AI bubble is predicted to collapse soon, with declining investor enthusiasm and funding, potentially leading to failures of high-valued companies by the end of 2024.

Read original articleLink Icon
Why the collapse of the Generative AI bubble may be imminent

The imminent collapse of the Generative AI bubble is predicted to occur within days or weeks, rather than the previously suggested timeline of 2025. The surge in popularity of Generative AI began with the launch of ChatGPT in November 2022, which quickly attracted 100 million users. Despite this rapid growth, concerns have emerged regarding the actual effectiveness of Generative AI technologies, suggesting they may not perform as well as anticipated. The author, Gary Marcus, argues that while Generative AI will not disappear, investor enthusiasm is likely to wane, leading to a significant reduction in funding and potentially causing many high-valued companies to fail or be dismantled.

Marcus highlights his credentials, including a Ph.D. from MIT and experience in machine learning, to support his predictions. He notes that he has previously identified limitations in deep learning and made accurate forecasts about the shortcomings of models like GPT-4. The current landscape, characterized by inflated valuations and unrealistic expectations, is expected to shift dramatically by the end of 2024. As the excitement around Generative AI diminishes, the financial implications for companies in this sector could be severe, with many unable to meet the high expectations set in recent years.

Link Icon 6 comments
By @bentt - 4 months
All those “bona fides” were self assessed. Pass.

And the thing is, I am not in disagreeement that the hype has gotten ahead of the reality. But the rationale for a collapse has more to do with human and market dynamics than “LLMs hallucinate”. The web was All That but we still had a bubble in 2000.

By @gsibble - 4 months
Agreed. Also been calling it for a while. Despite my best attempts, the only real uses for LLMs I've found are making chatbots (especially to interact with APIs), writing things like press releases, and using Github Copilot (which is admittedly awesome).
By @Havoc - 4 months
Sales pitch with zero value for a substack
By @thebeardisred - 4 months
It's not exactly throwing your hat over the wall to make a bold statement and then paywall the justification. His "boba fides" are "I did something 12 years ago, also my other predictions weren't wrong".

I like a positive track records but "not being wrong" isn't the same as being a true prognosticator.

By @nilawafer - 4 months
Paywalled