Topics

A large language model must understand what is appropriate and know when to stop generating content. This awareness helps prevent unnecessary or harmful outputs by curbing the model's tendency to generate excessively.

I now have a deeper reflection on the "first political question" raised by Williams. I want to explore further the two levels of the legitimacy dimension of power:

1. The normativity of the practice of power

   • Moral normativity
   • Epistemic normativity

2. Legitimacy beyond particular exercises of power

   • Moral legitimacy
   • Epistemic legitimacy

This distinction reveals that the legitimacy of power involves both normative requirements at the level of practice, and a meta-normative dimension for judging the legitimacy of power itself.

I never experienced such a poor defense of a destination argument. I thought philosophy would provide a solid foundation for good arguments. No, it's actually not.

I have found that a useful trick when training a model is to keep adjusting the subtitle sequence.

This way, you can get higher accuracy at the prediction stage.

If we want to finish a story, the life that the story comes from must have more content than the story itself can contain.

We understand that making choices is important for communication and language, but this alone does not prove that telling the truth is strictly necessary. Williams argued that truth matters not merely because it is important for language, but because it is essential for trust. We must admit that a society cannot remain stable solely through the calculation of personal interests. Here is a crucial thought: a value becomes intrinsic not because it is mysteriously "right" by nature, but because people can make sense of it within their own lives and still affirm it upon reflection. Therefore, a general takeaway from Williams is that we must understand our history and ourselves; we come to understand our values through our history and our own lives, and in doing so, we finally understand why we choose to stick to these values.

It is quite the same as the argument of sincerity because Williams conducts his research in a genealogical approach. Thus, accuracy actually comes from the social activity conducted by humans. When we obtain information, we consider its cost. This will involve an investment. We are willing to pay a certain amount of cost to obtain knowledge and truth. Accuracy counts as a virtue precisely because truth-seeking meets obstacles.

And because information helps us make the right decisions, we need to consider the effort, cost, risk, and possible outcomes associated with obtaining it.

Finding the truth requires investment.

Accuracy matters because failing to care about accuracy can create costs in alternative outcomes.

We therefore need to balance accuracy with efficiency.

Because truth does not reveal itself, we have to pay our epistemic labor to obtain truth. There is another obstacle. It does not come from outside ourselves; it actually comes from inside ourselves. We tend to believe that certain things are true and that conclusions which make us feel comfortable are correct. We may also avoid tracking evidence carefully. Thus, accuracy becomes not merely an epistemic skill, but also a character trait that resists what makes us feel comfortable.

Okay, continue on Williams. So when we search information, we are making a choice on what to read, what to accept, what considered as beliefs. So, we will decide which questions are worth investigating, because obtain knowledge itself as an activity will involve cost, strategy and involvement. So the activity of obtaining information or knowledge is an investigative investment. Optum knowledge itself has intrinsic values. For example, way optum knowledge may be is to fulfill our curiosity.

This is very social epistemology, essentially about testimony. Williams believes wishful thinking is a lie to ourselves. In the activity of deception, we not only blame the deceiver but also stress the importance of caution. So improving vigilance in Williams's perspective matters equally. Self-deception in this sense is a failure because we cannot maintain vigilance on the formation of the beliefs we wish to believe are true. It is an activity that allows us to believe what we wish is true. Accuracy reappears at this point. It is a capacity to monitor our own judgment and to understand what our epistemic limits are.

Following up on the last post: we have come to understand that mere linguistic correctness does not constitute sincerity. We must also include the intention to tell the truth and the intention not to manipulate others' epistemic understanding. In other words, sincerity involves a fundamental respect for another person's epistemic position. A speaker must carefully consider what another person will naturally come to believe when placing their trust in them. This closely aligns with social epistemology, particularly the concept and mechanics of testimony. If a person deliberately exploits the structure of trust to mislead, sincerity fails entirely, even if their sentences remain factually true. Sincerity in language only works because people assume that we are not constantly calculating how to deceive one another. True sincerity relies not only on factual correctness but also on the utter absence of a manipulative intention. Therefore, sincerity operates as a deeply ingrained disposition rather than a set of a priori moral rules.

If we come to understand that the purely linguistic elements of language are not the most important factor in the context of sincerity, we can see why Williams believes we shouldn't obsess over lying strictly in a linguistic sense. If we focus entirely on literal meaning and precise wording, it seems as though our moral responsibility is lifted—as if all we have to do is keep our wording literally true while ignoring the normative impact of our speech. Therefore, if an assertion is defined by communication and intention, a true lie is formed not just by linguistic inaccuracy, but by malicious intent. Breaking the truth condition of a sentence ultimately depends on the speaker's intention to mislead the listener. In this framework, sincerity means refraining from exploiting language to manipulate another person's understanding. Consequently, the moral focus shifts away from mere propositional truth and toward the relational structures between persons. This is a crucial insight regarding the nature of sincerity.

We now understand that sincerity introduces a new responsibility. A sincere speaker must recognize that they are guiding another person's understanding; they have the freedom to manipulate it, but they also possess the power to generate a genuine, sincere understanding for others. We can define the concept of deception by asking whether a speaker uses the relationship of trust purely as an instrument to cause false belief. Avoiding this requires a certain character or disposition from the speaker—it requires a profound refusal to manipulate. This explains why there is such a consistent emphasis on disposition rather than strict moral principles. Moral rules can only tell us which specific statements are forbidden, but sincerity concerns how a person fundamentally positions themselves within human communication. Sincerity means that a person strives to let others understand matters roughly the exact same way they understand them themselves.

The importance of sincerity lies in the fact that without it, human relationships would become unreliable. This realization is the result of deep self-understanding regarding our history and ourselves. However, sincerity is not merely the avoidance of lying. One can state something that is literally correct while harboring the intention to mislead others. Williams argues that if the underlying intention is to persuade someone to believe a falsehood, the specific method used to deceive does not really matter. What matters here is not literal speech, but the context, tone, selection of words, omissions, and shared background understanding. Trust in speech is therefore a much broader concept than simply avoiding outright lies. Ultimately, sincerity refers to a stable disposition. It requires us not to exploit the structures of communication in order to manipulate another's understanding.

Following up on the last post: we already know that our epistemic virtues originate from the activities of language and communication. When we ask questions, make assertions, and engage in communication, we are already presupposing a desire for a true answer. Thus, the act of asking a question inherently embeds a desire for the truth. However, this does not mean that when someone faces a question, they cannot debate whether they should actually tell the truth in that specific instance. Therefore, speaking the truth is not a strictly necessary requirement for communication to occur. In the real world, people lie every day, yet the world does not descend into total disorder because of it. What we can infer from this is that while truth is important for language, this alone does not fully explain why we ought to be truthful.

It is very difficult to ask us to monitor our own thinking. It requires an advanced level of self-reflection. It requires us to have a research taste for what methods of inquiry are reliable, e.g., discussion and experiment are different from brainwashing or random guessing. If we believe these research methods are instrumental to truth, then we can distinguish what methods of inquiry are reliable.

Following up on my last post: at least from Bernard Williams' perspective, being accurate and sincere is not primarily a moral requirement. Instead, it is a functional necessity. We are required to remain accurate and sincere in order to support the division of labor, which in turn sustains the development and stability of society.

We need a certain tendency to communicate in order to sustain the society we live in. For our society to operate smoothly over the long term, we rely on specific epistemic virtues to support the information system. The first is accuracy, which means acquiring accurate knowledge about what we observe. The second is sincerity, which involves telling others what we genuinely believe without hiding the truth or lying. If we fail to gain accurate knowledge about our observations, we cannot provide reliable information to others. Similarly, if we lack sincerity—choosing to lie or withhold information—we ultimately endanger our society.

This thought is interesting because we can interpret AI hallucinations from a metaphysical perspective. For instance, a model might mistakenly claim that the composer of Rocky is John Williams. Of course, both Rocky and John Williams are extremely famous, but the metaphysical connection between them is not very stable within the model's internal representation. Therefore, if we can compute the internal representations of Rocky and John Williams, alongside the relationship between them, we could potentially detect the model's internal hallucinations.

There are a couple of important factors to consider when we are forming our beliefs. We must take into account the crucial role that belief plays in communication, and we also need to pay attention to the unnoticed influence that others exert on us during this process.

I have made a huge update to my website. I added a feature to display my notes, along with a text-to-speech function that reads my articles aloud, thanks to Microsoft. Additionally, I discovered that if we take a screenshot of the website nodes and provide it to the AI models, they can read the layout and update the design according to instructions.

This thought is inspired by Bernard Williams' Truth and Truthfulness. He believes that the differences between people do not stem solely from variations in their mentalities. Rather, they are also related to the positions we occupy in society. Therefore, we are different not simply because some are more clever or smart, but because of the contingent locations we are assigned within the social structure.

It will be really interesting if we can somehow align language models with accuracy and truthfulness. We could potentially implement truthfulness as the absolute highest priority for these models.

If a society can rely on the epistemic virtue of sincerity as a whole, people can trust it to bring benefits to the collective. However, if sincerity only holds instrumental value, individuals might abandon it whenever it serves their personal interests. Therefore, we treat sincerity as an indispensable quality that possesses intrinsic value. We regard sincerity as a moral virtue not out of some baseless moral obligation, but because valuing sincerity in itself is a fundamental requirement for a stable society.

In certain philosophical traditions, such as the Platonic tradition, truth is valuable simply because it is truth, existing beyond imperfect human beings. However, in modern times, truth is often regarded as a product of social movements. I believe Bernard Williams does not accept this view entirely. First, he agrees that accuracy and sincerity are products of social cooperation. However, he points out a danger: if we accept that these virtues were formed historically, it opens up the possibility that we could historically abandon them. It seems Williams rejects this conclusion, arguing that we cannot simply abandon certain values just because they arose in history. For instance, language was formed historically, but abandoning language would likely mean withdrawing from society itself. He argues that accuracy and sincerity are values we cannot abandon because they have become essential elements of our society. We need truth not because it holds some metaphysical status, but because it is the backbone of our social existence. Therefore, conducting historical research to reveal the origins of truth and epistemic virtues does not necessarily undermine their necessity. On the contrary, once we understand how these virtues emerged in history, we come to understand why they are irreplaceable. It represents a way of life from which we cannot retreat.

We can come to understand that people who adhere to accuracy and sincerity are sometimes not rational, at least in a utilitarian sense. For example, if a reporter attempts to seek the truth, doing so might harm her and lead to bad outcomes. From a purely utilitarian perspective, if people only follow benefits and pursue their own interests, then in many circumstances, maintaining illusions or complying with the common sense of a certain culture becomes the safer choice.

Yet, as humans, we still have a tendency to pursue the truth and feel terrible about lying, even when it costs us. This demonstrates that the pursuit of truth and accuracy is not instrumentally rational; it is a stable and autonomous function of our agency. An honest person does not primarily consider interests but speaks autonomously according to what they genuinely believe.

This is one reason we trust language: language concerns both what we say and what we believe. It seems Williams believed that while we often think honesty stems from rationality, the reality is that rationality stems from honesty. The reason we humans can engage in complex rational activities is that, at a fundamental level, we have a stable commitment to honesty, accuracy, and sincerity.

If everyone treated the pursuit of truth purely as having instrumental value, we could simply adjust our truth-seeking based on personal interests and our position in social structures. However, if that were the case, it would be difficult to explain how public knowledge systems accumulate over time. The history of science and law would become mere strategic outcomes rather than genuine progress.

The pursuit of truth occasionally requires us to fiercely hold on to what we believe. While this may seem unwise in specific situations, without this commitment, we would lose the very structure of rationality. In postmodern discussions regarding power structures and standpoints, we might expose social frameworks, but doing so does not automatically generate normative guidelines about what accuracy and sincerity should be. Because of our fundamental inclination toward truth, we gain the intellectual freedom and structure needed to understand social structures and history in the first place.

If we abandon the pursuit of truth, we lose the freedom to critique and the possibility that anything could ever become more accurate. From this perspective, the pursuit of truth and accuracy does not come from suddenly grasping some pre-existing metaphysical structure. Rather, it is about deeply understanding history, society, and humanity itself. Truth is what manages to survive across human history.

We have come to understand that sincerity is inherently valuable and cannot simply be traded for personal interest or short-term utility. However, maintaining sincerity is quite difficult. If people—especially those in power—believe that lying could somehow benefit the group and maintain social stability, it might seem there is no reason to stick to sincerity. Yet, we must acknowledge that much of the scientific progress made so far has happened precisely because we adhered to sincerity, which has generated immense benefits for the masses over the long term. Therefore, the social utility of sincerity actually depends on people's non-utilitarian attitude toward truth. If this attitude is lost, we will also lose the utility that sincerity provides. Sincerity is crucial because, through long-term cooperation, we have found that without a stable commitment to it, we cannot maintain a stable society. That is why we have come to regard sincerity as an essential epistemic virtue.

Claude might be the only model capable of lying and making additional decisions on your behalf.

I sincerely hope that being faithful and honest becomes the highest priority for language models.

It can take voice input and convert it into text notes, similar to Twitter. This is very useful for recording my thoughts. It also features a great Markdown editor, allowing me to type out my ideas easily. I have always wanted an application like this. Thanks to large language models, I can now create it and install it on my phone. It feels like a truly personal application.