Is AI in a Bubble?

Rich Gilbert

1,368 views • 15 days ago

Video Summary

The speaker asserts that we are currently in an AI bubble, emphasizing that this doesn't preclude making money or advancing one's career within the technology. Drawing from years of experience building AI systems and implementing them in corporate settings, the speaker breaks down the AI phenomenon into technical, business/financial, and hype cycle aspects. While acknowledging the excitement surrounding AI, particularly after the GPT-3 moment, they caution against the sensationalized narratives often promoted, which can fuel fear and investment.

The core of the AI models, large language models (LLMs), are described as prediction engines that primarily predict the next word based on vast internet training data. The speaker argues that these models cannot generate truly novel concepts and that their "hallucinations" are simply errors stemming from the training data, which lacks an "I don't know" response. While LLMs are useful for tasks like summarizing documents or coding, they are compared to tools like spreadsheets, suggesting they won't wholesale eliminate jobs but may make some roles more challenging.

Recent studies, including one from MIT, indicate that a significant percentage of corporations implementing generative AI are not seeing profitability, highlighting a disconnect between the hype and the reality of its corporate value. The speaker likens this to past economic bubbles, noting that they tend to persist longer than anticipated and pop when even skeptics start to believe the hype. Despite this, the period before a bubble pops offers substantial opportunities for implementation and career growth within AI.

Short Highlights

The current AI landscape is characterized as an economic and hype bubble, though not necessarily short-lived.
Large Language Models (LLMs) function as prediction engines trained on internet data, capable of "hallucinating" when lacking information.
An MIT study reveals that 95% of corporations implementing generative AI are not profitable, indicating a gap between hype and corporate value.
Past economic bubbles, like the dot-com and housing bubbles, show a pattern of persistence and a tendency to pop when skepticism wanes.
Despite the bubble, significant opportunities exist for career advancement and implementation within AI before its eventual pop.

Key Details

The Nature of Bubbles and AI [0:00]

The discussion begins by defining economic, financial, product, and hype bubbles and how they operate.
The speaker, with extensive experience in AI development and corporate implementation, believes we are currently in an AI bubble.
This bubble doesn't mean an immediate pop or an end to profit-making opportunities within AI.

We are absolutely in an AI bubble. It doesn't mean it's going to pop anytime soon. It doesn't mean that there's not money to be made.

AI's Technical and Hype Dimensions [1:00]

The speaker shares personal history working with AI since the late 1990s, including building TensorFlow models at Google.
The current AI landscape is viewed through three lenses: technical, business/financial, and the hype cycle.
Initial excitement over GPT-3 led to predictions of significant job displacement, but this perspective has since been re-evaluated.

what what are we doing so there's an MIT study that came out we'll talk all about that today

Understanding Large Language Models (LLMs) [5:00]

LLMs are primarily prediction engines that predict the next word based on their training data.
They are trained on the internet, which "never says, 'I don't know'," leading LLMs to generate answers even when they lack information.
This lack of knowing, or "hallucination," is an error, but often unacknowledged to maintain the hype cycle.

on a technical basis, it does what every other machine learning algorithm does, right? Every deep neuronet network.

And you, you know, already I'm sure people have clicked off, but this is the truth. This is it can only say what it uh what it's seen.

The Disconnect Between Hype and Corporate Reality [10:20]

A recent MIT study indicated that 95% of corporations attempting to implement generative AI are not profitable.
This highlights a mismatch between the intense AI hype and its actual value creation in enterprises.
The consultants implementing these systems are profiting, while the corporations themselves often do not see tangible returns.

So people are starting to realize this. MIT came out with a study a few weeks ago, I think in August or or July um uh 2025 saying 95% of corporations that are trying to implement generative AI that are that have a a program like active and funded are not profitable are not making any success there.

The Patterns of Bubbles and Their Inevitable Pop [13:40]

The speaker compares the current AI bubble to past ones like the internet and housing bubbles.
A key characteristic of bubbles is the uncertainty regarding their actual size (hype vs. reality) and timing of the pop.
Bubbles tend to persist, and the moment naysayers begin to concede it's not a bubble is often when it's about to burst.

The problem with bubbles is nobody knows how big it actually is right like the the bubble is the the hype versus the reality and nobody knows how big that hype nobody can agree on the hype versus reality

That's when the bubble pops. That's always when the bubble pops.