AI is Overhyped When Will the Bubble Burst?

A bubble doesn’t mean something is useless, fake, or will disappear - it means market valuation is above its real value. Eventually, the market corrects and value drops.

The debate isn’t whether AI is useful. It clearly is, at least for some groups like programmers. The question is whether it’s overvalued.

While AI has impressive capabilities, it’s far from revolutionizing the world or achieving the total automation the public expects in the short term. AGI that could replace all employees is still far off, if it’s even possible. The closest analogy is Searle’s Chinese Room .

I’m not alone in this view. Apple stepped back from parts of the AI race, with reasons outlined in The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity .

Sam Altman has also warned about an AI bubble :

“Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes. Is AI the most important thing to happen in a very long time? My opinion is also yes,”

Many AI companies can’t live up to expectations

Builder.AI, which marketed AI-powered website creation, turned out to rely heavily on Indian developers working behind the scenes and declared bankruptcy . That highlights the gap between marketing and reality.

Many AI companies raise money by creating expectations, then struggle when they can’t deliver. By the time that happens, they’ve often already raised significant funding.

Devin AI also faded into obscurity. We’re seeing diminishing returns with each new LLM release as generative models approach a plateau.

Many companies have shifted away from the “replace developers” narrative toward becoming code creation or assistant tools, like Bolt, Lovable, or V0 , along with tools like OpenCode, Claude Code, and Codex.

Humanity’s Last Exam shows we’re still far from AGI

Humanity’s Last Exam scores sit around 30% - impressive, and a significant breakthrough. But that’s far from 100%, despite what some clickbait headlines suggest.

These results are notable, but they don’t mean we’re close to AGI as often portrayed.

The AI bubble is fueled by hype and misunderstanding

Many companies add “AI” to their pitch to attract attention from investors and customers. Most people don’t understand how these models work statistically, so they often attribute them with near-magical abilities.

Businesses hope AI will cut hiring costs, while investors hope to find the next big AI winner.

Financial frenzy

Investors are rushing to fund anything labeled “AI” and “disruptive”, hoping to catch the next big winner.

Much of this mirrors past tech cycles. The market is trying to profit from sudden interest in something abstract like AI.

AI-driven layoffs

Layoffs have created fear of replacement, but the bar is rising for different reasons. Most layoffs aren’t directly driven by AI - companies are using AI as a cover to justify cost cuts .

Many AI solutions are just ChatGPT wrappers

Many startups are just thin wrappers around ChatGPT or other LLMs with a nicer UI, hoping to become the next unicorn. If your product is just a wrapper, it’s easy to replicate and hard to defend.

This isn’t inherently bad, but it inflates valuations. Protocols like the Model Context Protocol make building these wrappers even easier.

I’d rather see more investment going toward applications that actually add value - like using AI to design better drugs or automating tedious work to free up time, not just chasing quick returns.

Running out of training data

AI is limited by data and compute. Compute can scale, but high-quality training data is getting scarce. Over 50% of online content is now AI-generated , creating a feedback loop where models train on AI-generated content and returns diminish.

The end of the AI bubble

AI isn’t smoke and mirrors - it’s here to stay. But its social impact in the short term will likely be concentrated among a few large players like OpenAI, Google, and Microsoft.

At its core, current AI is advanced autocomplete for text, images, and video. It’s useful for translation, though relying on it for live conversations has drawbacks . AGI remains distant.

Machines don’t learn. What a typical “learning machine” does is finding a mathematical formula, which, when applied to a collection of inputs (called “training data”), produces the desired outputs. This mathematical formula also generates the correct outputs for most other inputs (distinct from the training data) on the condition that those inputs come from the same or a similar statistical distribution as the one the training data was drawn from. - Andriy Burkov

The financial bubble will burst and hype will fade. We’ll reach a point where people understand AI’s real capabilities and limits, and it becomes just another tool. The internet will change, and the AI FOMO will move to the next trend.