With the surge in popularity of DeepSeek, the debate on whether "AI will render liberal arts obsolete" has resurfaced, with related topics trending on social media.

Many contend that large models like DeepSeek can create high-quality humanities works and access vast humanities knowledge online. AI is already adept at writing official documents, academic papers, and marketing copy. Additionally, there are large models for text-to-image, text-to-video, and text-to-music generation, potentially replacing artistic and design work.

At first glance, it might seem that AI will soon replace liberal arts jobs, making a career in humanities seem like a dead-end choice. This public opinion pressure has spread fear among liberal arts students and their parents, casting a shadow over humanities education in the AI-dominated era.

But is this really the case?

Each time AI technology gains traction, there are those who revisit the "liberal arts obsolescence theory." However, the AI excitement sparked by DeepSeek precisely underscores the importance of humanities capabilities in the AI race, and even the employment prospects for liberal arts students in the AI era.

Consider this: If AI makes liberal arts obsolete, do large AI models still need liberal arts students?

A few examples highlight that humanities capabilities are crucial in today's large model competition, even somewhat determining its direction.

Take DeepSeek, for instance. Comparing it with other large models, DeepSeek doesn't significantly differ in basic capabilities like information gathering and even has hallucination issues. Yet, it has continued to explode in popularity globally. Besides factors like reduced AI hardware costs, the most significant reason is its exceptional topic spreadability.

DeepSeek's answers resonate better with internet culture, possess a stronger sense of humor, and are more elegant and poetic. These abilities make DeepSeek's responses feel more "human" when addressing the same question, piquing users' interest in conversing with it and sharing their interactions on social media.

So, do these elements of humor, wit, and poetry stem from algorithmic optimization? Likely not. At least, the meticulous selection of training data for the model plays a pivotal role. This grasp of literary quality, philosophical thinking, sense of humor, and even social spreadability arguably falls within the domain of liberal arts students.

An example that underscores DeepSeek's emphasis on humanities capabilities and the support of liberal arts students is its response to macro-level questions. It often responds with grandiose, profound-sounding words lacking practical meaning, like "gene cube" and "phantom cage." This resembles the tactic of making up nonsense when humanities students struggle to write their theses, mainly aimed at sounding impressive but being useless and untraceable by plagiarism checkers.

Looking back at DeepSeek's rapid rise, it becomes evident that its emphasis on humanities capabilities sets it apart. The so-called "hundred models war, all the same." One reason for the severe homogenization of large models in the past lies in the overemphasis on algorithms and neglect of content, ignoring that AIGC is essentially content creation. Recently, the AI industry has been discussing ways to reduce training and inference costs, similar to what DeepSeek has achieved. However, without recognizing its emphasis on humanities capabilities and content generation, most subsequent large models will still fail to escape the vortex of boring content, uninteresting dialogues, and eventual abandonment by users.

Here are more examples highlighting the importance of humanities capabilities to large models. I've noticed that friends in civil servant and teacher-related professions prefer to use ERNIE Bot for writing official documents, especially work summaries and meeting minutes. Their reason is that ERNIE Bot's language is more appropriate, and its format is neater compared to other large models. This is crucial in public service work and can reduce subsequent revision efforts. From this perspective, understanding liberal arts-oriented tasks like document writing has become key for large models to target user groups and enhance their competitiveness.

Another example can be seen in text-to-image large models. To this day, we still marvel at the lack of a domestic alternative to Midjourney. Despite its difficult-to-use interface, complex prompts, and laggy network, users with specialized needs continue to flock to Midjourney. The root cause lies in the poor aesthetics of a large number of domestic text-to-image models, resulting in generated images that exude an indescribable tackiness. The reason behind this is the lack of high-quality paintings and design works for AI to learn from. Artistic, design, and aesthetic abilities, which are liberal arts skills, have severely limited the development of text-to-image, text-to-video, and even the entire multimodal route.

Who says liberal arts can't determine AI productivity?

To some extent, AI will not render liberal arts useless. Without liberal arts students, AI might become useless first.

Actually, the argument that "when a certain advanced technology arrives, liberal arts become useless" is already cliché. Remember when the internet first emerged, some said that liberal arts knowledge would be readily available with a search, so what's the point of rote memorization in liberal arts? Later, when self-media emerged, there were those who said anyone could write for self-media, so what future do liberal arts students working for newspapers and magazines have?

The subsequent story was quite the opposite. The internet industry and self-media became strong growth engines for liberal arts employment. Even some liberal arts majors with narrow employment prospects, such as archival science and museology, found new career paths through science-popularization-oriented self-media. It was said that with the advent of short videos, no one would read text anymore. However, jobs like short drama screenwriters and video planners have also absorbed a large number of liberal arts graduates. As technology advances, liberal arts students not only have a wider range of employment options but also a relatively greater possibility of entering high-income brackets.

Today, it seems that large models are again coming to squeeze out the jobs of liberal arts students. But we easily overlook the fact that the content of large models doesn't emerge from thin air; it also requires humanities workers to provide training data and supervise effective content generation. On the other hand, large models can also serve as new content creation assistants and workplace aids, similar to self-media platforms.

Even in the initial stages of large model proliferation, we can already see a plethora of new liberal arts employment directions emerging, such as:

1. AIGC Content Officers behind large models.

DeepSeek's success has already shown that the quality of content generation will be a key factor determining the success or failure of large models. Next, we are bound to see a new trend of AI vendors placing greater emphasis on humanities capabilities and the selection of high-quality training data. In this process, the task of training data screening, which could originally be outsourced at a low cost, will become a high-value job led by professionals with a humanities background. The role of AI Content Officer is probably not far off.

2. Intelligent Agent Developers.

What is the biggest obstacle for liberal arts students to become software application developers? It should be the inability to program. Well, now AI can do it. While large models are lowering the threshold for content generation, they are also eliminating the thresholds for technical roles like coding. Application development driven by a humanities background is likely to become a new employment and entrepreneurship boom.

3. Multimodal Content Creators.

Similarly, it's challenging for a liberal arts student to transform knowledge, theories, and literary ideas into images. This requires crossing over into the artistic field or even acquiring technical skills like video editing, but AI can handle these tasks. After text, images, and videos, the next windfall for liberal arts employment might be multimodal creators.

Summarizing these new opportunities, we may reach a consensus: AI brings not disciplinary discrimination but opportunity equity.

Whether it's filling out invoices, factory quality inspection, writing code, generating copy, or writing reports and summaries, AI can only replace simple, repetitive tasks for a long time from now to the future. Therefore, the jobs it can replace are concentrated in this area—those that are simple and highly repetitive.

At the same time, the fact that AI can do simple tasks also means it can complement everyone's skill deficiencies. Pure coders can rely on large models to acquire systematic knowledge of the humanities and social sciences; those who have never been exposed to artistic creation can use AIGC to enhance their aesthetic and design capabilities; and an entrepreneur with a purely liberal arts background can also become an AI developer by relying on intelligent agents and no-code development tools, thus riding the wave of the times.

The biggest impact of the AI era on individuals is not that some disciplines are useful while others are not, but that it requires us to reorganize our skill sets: Each person must possess creative abilities that cannot be replaced by AI and learn to delegate tasks they are not good at or lack interest in to automation and intelligence, thereby achieving more comprehensive competitiveness and fairer social opportunities.

Upgrades in technological tools have always led the world towards openness and inclusivity. From this perspective, intelligence equals equity.