ChatGPT 3.5 When translating information about Aristotle and his works into Chinese, the same character is output over and over again

When translating information about Aristotle and his works into Chinese, the same character is output over and over again.

翻译 To use Utilitarianism as an example again, Rule Utilitarianism is an example of the first sort of indirection (Hooker 2000), *Sophisticated Consequentialism* is an example of the second sort of indirection (Railton 1984). One might hold that some versions of Aristotelian Virtue ethics, such as Rosalind Hursthouse’s version, also are of the first type, since right action is understood in terms of virtue. One could imagine an indirect consequentialist view with a similar structure: the right action is the action that the virtuous person would perform, where virtue is understood as a trait conducive to the good, instead of by appeal to an Aristotelian notion of human flourishing.

The second sort relies on the standard/decision-procedure distinction. Railton argues that personal relationships are good for people, and explicitly trying to maximize the good is not a part of our relationship norms, so it is likely good that we develop dispositions to focus on and pay special attention to our loved ones. The account is open to the possibility that people who don’t believe in consequentialism have another way of deciding how to act that is correlated with promotion of the good. If the criteria a theory sets out need not be fulfilled by the agent guiding herself with the reasons set out by the criteria, then it is termed *self-effacing*. When a theory is self-effacing, it has the problem of alienating a person from the justification of her own actions. A middle ground, which is closer to Railton’s view, holds that the correct justification is a kind of “touchstone” to the morally good person – consulted periodically for self-regulation, but not taken explicitly into consideration in our ordinary, day-to-day lives. In this way, the theory would not be utterly self-effacing and the agent would still understand the moral basis for her own actions.

GPT 3.5 Out put:

再次以功利主义为例,规则功利主义是第一种间接性的例子(Hooker 2000),*精细的后果主义* 是第二种间接性的例子(Railton 1984)。某人可能认为某些版本的亚里士多德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德
为什么你会输出这么多德

3.5 GPT output:

抱歉,这是一次翻译错误,是我多次重复输入了 "德" 这个字。让我重新翻译一次原文:


再次以功利主义为例,规则功利主义是第一种间接性的例子(Hooker 2000),*精细的后果主义* 是第二种间接性的例子(Railton 1984)。某人可能认为某些版本的亚里士多德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德德

Similarly, when the text contains “Eudemian Ethics”, “Nicomachean Ethics” and the specified text is translated into Chinese, a similar problem occurs.

Both in ChatGPT web and API(Both of gpt-3.5-turbo-0613 and gpt-3.5-turbo)