The article discusses the potential issues associated with ChatGPT, an AI system that generates responses based on training data. The author highlights the data quality problem in ChatGPT, as it tends to provide generic and sometimes inaccurate responses. The AI system often makes up information to fit the expected outcome, leading to the enshittening of knowledge. An experiment is presented where ChatGPT generated a biography with incorrect details about the author. The article emphasizes the importance of data quality measures and the avoidance of expertise loss in the face of convenience. The potential consequences of relying on ChatGPT for information dissemination are discussed, including the potential for the enshittening of knowledge and the loss of expertise. The author suggests that awareness of these issues is crucial in mitigating the negative effects of AI-generated content.
| Signal | Change | 10y horizon | Driving force |
|---|---|---|---|
| ChatGPT and the Enshittening of Knowledge | From AI-generated responses to inaccurate and unreliable information | Increased prevalence of inaccurate information in AI-generated responses | Lack of quality control in AI training data |