Sign up to save your podcasts
Or
Share Did the AI bubble just Burst?
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Did the AI bubble just Burst?
Did the AI Bubble Just Pop?
Hey, so the question on everybody's mind today is, did the AI bubble just pop? I think no. The stock market thinks yes. Let's explore why I believe that perspective is a bit short-sighted and doesn't consider the bigger picture.
The Expanding Role of AI
Here's how I use AI in 2025 for tasks that weren't possible just 24 months ago:
Two years ago, these applications of AI were not possible because the models simply didn't exist. We're only a couple of years into this transformation, and the potential applications of AI are still largely untapped.
Limitations and Future Potential
Currently, AI can't handle tasks like doing taxes or providing legal advice. The training data exists—case law is just a collection of books that someone spends years understanding. While AI might not take over the legal industry, it could make knowledge of tax law or other legal fields more accessible through well-trained models, but we're not there yet.
Demand for AI Remains Strong
The demand for AI hasn't changed. Despite headlines claiming that a model was trained for $6 million to compete with OpenAI's cutting-edge model, I believe that's mostly propaganda. Significant advancements include activating fewer parameters per token and using a mixture of experts. In the coming days, OpenAI is expected to disclose how GPT-4 Turbo was made cheaper and more resource-efficient than previous versions.
Infrastructure Challenges
Building a data center to run these models still requires hundreds of millions of dollars. It's not like new models will suddenly run on different chips. They will continue to operate on H100s in data centers that require significant electricity and human resources to build and maintain. We are still not meeting the demand for AI infrastructure, as more people want to use tools like Stable Diffusion or ChatGPT than there is capacity for.
Conclusion
The assertion that AI demand has decreased is misleading. There are still new applications for AI models, and the infrastructure to support them is continually evolving. The claim that a model was trained for $6 million on half a billion dollars worth of GPUs is part of a larger story about the ongoing development and deployment of AI technologies.
View all episodes
By Indie.am
5
11 ratings
Did the AI Bubble Just Pop?
Hey, so the question on everybody's mind today is, did the AI bubble just pop? I think no. The stock market thinks yes. Let's explore why I believe that perspective is a bit short-sighted and doesn't consider the bigger picture.
The Expanding Role of AI
Here's how I use AI in 2025 for tasks that weren't possible just 24 months ago:
Two years ago, these applications of AI were not possible because the models simply didn't exist. We're only a couple of years into this transformation, and the potential applications of AI are still largely untapped.
Limitations and Future Potential
Currently, AI can't handle tasks like doing taxes or providing legal advice. The training data exists—case law is just a collection of books that someone spends years understanding. While AI might not take over the legal industry, it could make knowledge of tax law or other legal fields more accessible through well-trained models, but we're not there yet.
Demand for AI Remains Strong
The demand for AI hasn't changed. Despite headlines claiming that a model was trained for $6 million to compete with OpenAI's cutting-edge model, I believe that's mostly propaganda. Significant advancements include activating fewer parameters per token and using a mixture of experts. In the coming days, OpenAI is expected to disclose how GPT-4 Turbo was made cheaper and more resource-efficient than previous versions.
Infrastructure Challenges
Building a data center to run these models still requires hundreds of millions of dollars. It's not like new models will suddenly run on different chips. They will continue to operate on H100s in data centers that require significant electricity and human resources to build and maintain. We are still not meeting the demand for AI infrastructure, as more people want to use tools like Stable Diffusion or ChatGPT than there is capacity for.
Conclusion
The assertion that AI demand has decreased is misleading. There are still new applications for AI models, and the infrastructure to support them is continually evolving. The claim that a model was trained for $6 million on half a billion dollars worth of GPUs is part of a larger story about the ongoing development and deployment of AI technologies.