Why It’s Extremely Hard to Start an AI Application Business with Large Language Models

🌈 Abstract

The article discusses the challenges of starting an AI application business based on large language models (LLMs). It argues that most startups focusing solely on LLM applications are likely to fail within six months due to the difficulty of finding genuine user needs, the unreliable performance of LLMs, and the strong competition from ChatGPT.

🙋 Q&A

[01] Why It's Extremely Hard to Start an AI Application Business with Large Language Models

1. What are the key challenges in starting an AI application business based on large language models?

• Most essential user needs are already well met in today's oversaturated market, making it hard to find genuine pain points that AI can address.
• AI solutions based on LLMs often fall short of expectations, with accuracy typically around 80-90%, resulting in only 64% reliability when multiple API calls are required.
• Users are increasingly impatient and expect a 10x improvement over existing solutions, which is difficult to achieve with the current 70-80% success rate of AI.
• ChatGPT offers greater flexibility and user trust, making it hard for new AI applications to compete.
• The high costs of using LLMs, due to extensive prompt engineering and token usage, make it challenging to generate a profitable business model.

2. Why are voice synthesis applications more successful than other AI modalities?

• Voice synthesis applications don't necessarily solve problems, but rather generate new traffic, which is more important in today's oversupplied market.
• Voice synthesis has the ability to create viral content, which is a key factor in its success.

[02] The Potential of Other AI Modalities

1. What other AI modalities does the author see potential in, besides large language models? The author sees a lot of potential in other AI modalities, such as images, videos, and 3D, rather than just focusing on large language models.