Sự nguy hiểm của việc quá tin tưởng AI

Tôi check một đoạn paper của mình dưới đây bằng GPTZero, và kết quả nó nhận định là "100% generated by AI":

The first cluster examines how tools, standards, and institutional arrangements are used to support social and environmental goals. Instead of finding tools replacing traditional finance, these studies search for methods to embed impact-oriented objectives within existing policy frameworks, market mechanisms, and stakeholder practices. A common topic in this group is the deployment of outcome-based instruments, particularly social impact bonds (SIBs), to improve service delivery by linking funding to measurable results. Dowling (2017) and Berndt and Wirth (2018) highlight that these instruments have changed performance evaluation and risk allocation in funding projects. A number of publications have pointed out the weaknesses of current indicators, arguing that they often fail to capture the complexity of real social changes (Arena et al., 2018; Reeder et al., 2016). At the same time, Nicholls (2010) and Bugg-Levine and Emerson (2011) emphasized that common reporting standards have played an important role in building the credibility of impact investing in mainstream financial systems. A third group of discussion in this cluster is the role of digital technology in sustainable finance. Moro-Visconti et al. (2020) have studied how data analytics and algorithm-based models could be used to integrate ESG factors into investment decisions. Although generally considered as a new method to increase transparency and efficiency, concerns have been raised about these tools in social programs, especially regarding accountability. This cluster shows that impact finance has continued to develop within the structures of existing institutions, but priority and measurement is still under argue between authors.

Tuy nhiên, khi tôi tách nó ra làm 3, thì kết quả như sau:

The first cluster examines how tools, standards, and institutional arrangements are used to support social and environmental goals. Instead of finding tools replacing traditional finance, these studies search for methods to embed impact-oriented objectives within existing policy frameworks, market mechanisms, and stakeholder practices. A common topic in this group is the deployment of outcome-based instruments, particularly social impact bonds (SIBs), to improve service delivery by linking funding to measurable results.  -> Phần này kết quả là 97% human

Dowling (2017) and Berndt and Wirth (2018) highlight that these instruments have changed performance evaluation and risk allocation in funding projects. A number of publications have pointed out the weaknesses of current indicators, arguing that they often fail to capture the complexity of real social changes (Arena et al., 2018; Reeder et al., 2016). At the same time, Nicholls (2010) and Bugg-Levine and Emerson (2011) emphasized that common reporting standards have played an important role in building the credibility of impact investing in mainstream financial systems. -> phần này 95% human

A third group of discussion in this cluster is the role of digital technology in sustainable finance. Moro-Visconti et al. (2020) have studied how data analytics and algorithm-based models could be used to integrate ESG factors into investment decisions. Although generally considered as a new method to increase transparency and efficiency, concerns have been raised about these tools in social programs, especially regarding accountability. This cluster shows that impact finance has continued to develop within the structures of existing institutions, but priority and measurement is still under argue between authors. -> phần này 98% human

Bạn có thể tự test đoạn mẫu trên với GPTZero để kiểm chứng.

Tất nhiên, mỗi khi chúng ta làm việc với AI, luôn có một bảo lưu là AI có thể gặp sai sót, nhưng liệu chúng ta có đủ sự bình tĩnh suy xét để nghi ngờ nhận định của trí tuệ nhân tạo khi nó khẳng định "đây 100% là kết quả AI tạo ra"? Hiệu ứng anchoring effect, có lẽ, sẽ ngăn chúng ta đưa ra một quyết định đủ tỉnh táo, và điều này hoàn toàn có thể sẽ dẫn tới những kết luận không ổn một chút nào!

(Hay mình cũng là một con AI mà tự mình ko nhận ra???)