News A/B: Comparing Perspectives Across Multilingual News

Thu, 03 Sep 2026 00:00:00 +0000

Most people only read news coverage from their own country, in their native language. On current-affairs topics where people in different countries care about very different things, we end up hearing only our own side.

Take the recently viral film Niu Lai (牛来) in China. Both chinese and western media love to compare its footage with AI-generated video. But here’s the interesting part: a good number of Chinese outlets treat AI as a yardstick for measuring the film’s quality, calling it “worse than what AI could make”; meanwhile, the overwhelming majority of English-language media see the film as “a relief against AI-flooded content.”

Behind these divergent perspectives are, naturally, different cultural attitudes toward AI technology. Some articles might extend the point here: “The West is hostile to AI, while China embraces it.” Reading that sentence, some people will nod, and some will shake their heads.

Whatever your personal view, “the West is hostile to AI, China embraces it” is just an opinion with no evidence behind it (yet). In the paragraph above, I too used all sorts of vague words to describe the Niu Lai phenomenon: “a good number of Chinese outlets,” “the overwhelming majority of English-language media.” So here’s the question: why should you believe me? Just because you happen to agree with the opinion underneath?

When we say “A is the same as / different from B,” what are we actually saying? news-ab.com is my small experiment at answering that question within a fairly narrow scope: automatic insights discovery, backed by statistical evidence.

News A/B is a website that automatically discovers patterns and compares viewpoints across publicly available news in multiple languages.

For a given news event, AI collects coverage from two groups of media (different countries, different languages) and uses its multilingual ability to produce semantically unified “question - answer” annotations. Statistical methods then automatically compute three kinds of patterns, reproducible to a certain degree, in how the two sides answer the same question:

Consensus: both sides’ top-voted answers agree
Divergence: both sides’ top-voted answers differ
Silence: one side’s attention/response rate on the question is near zero, while the other side’s is not

Only patterns that pass statistical testing are picked by the AI and written up into the final highlights page, and every important claim on the page can be traced back to the original coverage and its citation. AI still makes mistakes, but at least what it produces here is auditable.

Because every step that requires semantic understanding uses AI throughout the pipeline, and no cost optimization has been done yet, the time and cost of generating a single report are currently quite impractical (3-4 hours, and the equivalent of $100-150 in token usage; of course, I was using my subscription quota). That’s why there are only a handful of reports on the site right now.

Most of the insights the AI finds aren’t all that interesting, either. Whether it’s worth the time and cost depends entirely on your interests and needs. So I’m sharing it as an experimental open-source project (GitHub link; mostly Agent Skills). If there’s a news topic you’d like to compare, and you have some AI quota you can’t use up, let your own agent give it a try.

(As an aside, today’s life advice: while nudging or waiting on multiple agents to do work, make sure you have work of your own in hand. Without it, you’re a restless foreman pacing back and forth; with it, you’ll think, “what a great team.”)

Written in Chinese, translated by Claude Fable 5.1.

News-Ab on ash's Writing Archive

News A/B: Comparing Perspectives Across Multilingual News