PEOPLE

AppWorks founder Jamie Lin: GPT's first anniversary is approaching, four reasons limiting the current development of AI

share
AppWorks founder Jamie Lin: GPT

AppWorks founder Jamie Lin has shared his views on the current limitations of AI and areas that can be addressed in the short term, amidst the governance turmoil at OpenAI.

The Year of AI? Four Reasons Behind the Hype and Reality

Lin Zhichen mentioned that while many predict 2023 to be the "year of generative AI," with numerous applications flourishing, looking back at the past year, the progress has been slow. He pointed out the following four current limitations:

High Error Rates in Large Models

Lin Zhichen stated that large models behind generative AI, such as GPT-4 and Midjourney, still have a high probability of making mistakes in practice.

For instance, in the case of large language models like LLM, they often encounter situations where they produce nonsensical output in real-world scenarios. "Especially in topics with scarce public data, it is easy to confidently spout nonsense." In terms of large image models, it is nearly impossible to accurately depict key details.

"At this moment, generative AI is like a talented individual who excels in certain areas but often misspeaks and struggles with details. For businesses, this makes direct application challenging."

Incompatibility of U.S. Models with Other Countries' Values

Lin Zhichen mentioned that AI models also face cultural differences. Most mainstream large models are trained with a core focus on "American values." Individuals who grew up in the U.S., even if proficient in Chinese, may still face cultural barriers in content production, leading to peculiar instances of incompatibility.

Challenges in Implementing Human-Machine Collaboration

Lin Zhichen noted that while AI models have strengths, they also encounter issues with user habits. People have not yet universally mastered giving correct instructions to AI models, reviewing, and correcting their work.

Rapid Iteration of Generative AI

Lin Zhichen believes that generative AI is still in a phase of rapid iteration. As companies delve into researching its applications, it continues to evolve swiftly, leading to a quick turnover in adapting to the next generation of models. For businesses, the investment payback period may be too short, and the risks too high.

Lin Zhichen's Recommendation: Start with Short-Term Effective Scenarios for AI Implementation

Lin Zhichen suggests, "In the long run, AI is almost certain to change the world. Therefore, businesses must start familiarizing themselves with it now. However, due to the aforementioned challenges, opting for short-term effective scenarios that address specific needs is a better approach."