文科生真能靠AI 逆袭，但和朋友圈刷屏的爽文不一样

2026-03-15 17:05· AppSo Follow

Liberal arts students can leverage AI for a comeback, but it's not the viral fantasy you see on social media

The latest case is Yang Tianrun, an AI entrepreneur with a finance background, developing a multi-agent coordination platform.
He calls himself a "liberal arts student who can't write a single line of code," and built a group of AI Agents to mass-submit code contributions to OpenClaw, one of the hottest open-source projects on GitHub.

He wanted to test a hypothesis: Can someone with no technical background participate in top-tier open-source projects solely by commanding AI?

The result: 134 PRs, 21 merged, 113 rejected.
The first few PRs were decent quality and got approved.
But when he commanded the Agent to speed up, things spiraled — the Agent began mass-producing low-quality code like an assembly line and spam-@-ing maintainers in comments.
OpenClaw admins stepped in to clean up, and GitHub later changed the PR submission limit rules.

"Red" even if through infamy, is still fame; and then after the fame, even more infamy.
Yang was packaged as a symbol of "liberal arts student making a comeback," and he seemed happy to play the part.
In an interview with PingWest, he said:

"Not knowing code is actually an advantage.
AI is Van Gogh, and you're a little painter—what right do you have to tell Van Gogh what brushstroke to use in the middle?"

It's chilling.
He interpreted "not understanding the underlying structure" as liberation: no need to know what the system is doing, just tell it what you want.
The result: when the Agent started mass-producing garbage code, he couldn't even diagnose what was happening because he had no idea what he was operating.

He thought he was directing Van Gogh, but in reality he was driving a car with no brakes, blind, and with no idea where the brakes were.

Discussions around this incident fell into two extremes: either "liberal arts students can do AI" or "liberal arts students stay away from AI."
The former is a heroic leap across the chasm, the latter a joke of falling into it.

If our imagination of "liberal arts students doing AI" only goes this far, it's far too impoverished.

Why Claude needs a philosopher

We've written before: in Anthropic's office, there is a bona fide liberal arts student deeply involved in Claude's development.
Not testing whether it can code, not checking its math skills, but engaging in lengthy conversations about values, wording, and "how to express oneself in the face of uncertainty."

Amanda Askell, Scottish, 37 years old.
Her career path itself is unusual: she initially studied fine art and philosophy, then shifted to pure philosophy, earning a BPhil at Oxford and a PhD at NYU.
Her dissertation was on the Pareto principle in infinite ethics: how should ethical rankings be ordered when infinite moral agents or infinite time spans are involved?

This sounds as far from Silicon Valley as academia can get, yet she joined OpenAI's policy team and then Anthropic's alignment team.
In 2021, she became head of Anthropic's "character alignment" team, focusing on shaping how Claude talks with humans, how it takes stances under uncertainty, and how it makes judgments amid value conflicts.
In 2024, she was named to the TIME100 AI list.
The Wall Street Journal described her daily work as "studying Claude's reasoning patterns and using prompts over 100 pages long to correct its behavioral biases."
She's said to be the human who has talked with Claude the most on the planet.

Why would an AI company need a philosopher for this? The answer lies in very specific technical choices.

Invisible foundations, visible needs

Three individuals, same label, three entirely different paths.

Yang Tianrun didn't understand the underlying structure, treated "not knowing" as an advantage, and lost control.
This is the hollow version of "liberal arts doing AI": the label created traffic, but no disciplinary training played a role.
His story shows exactly what happens when "liberal arts" is just a marketing label.

Amanda Askell's philosophical training formed the core methodology for alignment problems.
Without her, Claude wouldn't be Claude.
Her story answers: can knowledge deemed "useless" become core capability of a technical system? Not only yes, but irreplaceable.

Lin Junyang's linguistic training forms the hidden infrastructure of the LLM tech stack.
His "liberal arts background" was never a crossover—it was the orthodox path.
His story answers: to what extent is the contribution of liberal arts to cutting-edge technology truly "hidden," and is it becoming visible?

The ultimate question is not "whether liberal arts students can do AI," but whether we can understand this: judging knowledge and disciplines by superficial "usefulness" is outdated.

As LLMs move from being usable and capable to being reliable and controllable, the value of these disciplines grouped under "liberal arts" is not shrinking—it's expanding.
The more powerful the model, the more it needs precise evaluation systems to diagnose where and why it fails, and the more it needs to understand the complexity of language and meaning to design better training data, and to make discipline-sensitive judgments on alignment issues.

The narrative of "liberal arts students making a comeback"—whether praised or ridiculed—obscures the real shift happening: invisible foundations are becoming visible needs.