Businesses That Just Solve Problems Via LLM Models Will Go Bankrupt

This video is about the overuse of large language models (LLMs) in business.

The speaker argues that while LLMs are powerful tools, they are expensive and not always the best solution for every problem. Businesses that rely solely on LLMs to solve problems will eventually go bankrupt because they are not cost-effective.

The speaker proposes a more cost-effective approach that involves using a combination of LLMs, classical algorithms, and machine learning algorithms. This approach involves carefully considering the most cost-effective way to solve a problem, rather than simply throwing an LLM at it.

The speaker believes that the ability to understand the cost-effectiveness of different AI models will be a highly sought-after skill in the future. In the video, the speaker uses an ETL tool (extract, transform, load) as an example. He argues that it would be unnecessarily expensive to use an LLM for the entire ETL process, when a classical algorithm could be used for most of the work. The LLM could be used for a specific function within the ETL process, reducing the overall cost.

The speaker concludes by stating that businesses that can effectively scope out problems and use the most cost-effective AI models will be the most successful.

[johnpienta]

This is an extremely interesting iteration of model collapse.

Using LLMs to make content which fills the Internet and then ends up in the next training set... You get only a few iterations before it's completely unintelligible. LLMs are a form of data compression!