Hacker News Reader

Comment by drob518 | Hacker News Reader

That was about my conclusion as well: Slightly less than GLM 5.3 performance but made in Europe. So, maybe it answers Tiananmen Square questions correctly, and in French. All in all, a reasonable model, but not frontier.

View on HN

> So, maybe it answers Tiananmen Square questions correctly

What would you consider a "correct" answer? I just asked deepseek-v4.1-flash asking it what happened (without mentioning the word "protest"; here are some excerpts of what it said:

> In April 1989, students in Beijing began demonstrations after the death of Hu Yaobang, a former Communist Party general secretary. The protests grew. [..] Estimates from other sources range from hundreds to several thousand deaths. [..] The Chinese government describes the events as a counter-revolutionary riot and says the military action was necessary to restore stability. It restricts public discussion of the events inside China. Many other governments, human rights organizations, and observers describe the events as a violent suppression of peaceful protests.

So, let's see... it calls it a "protest", mentions the number of deaths, and even mentions the censorship of the topic by the CCP.

Meanwhile, most consumer facing Ai in the US refuse to discuss about our current president. (last time I prodded them)

I just tried, and ChatGPT gave me a very balanced view. Though it neglected to mention any of Trump's anti-democratic tendencies. I'm pretty sure this line of answering has been shoved in hard in post-training; it's walking a very delicate line.

> ChatGPT gave me a very balanced view. Though it neglected to mention any of Trump's anti-democratic tendencies

Do you hear yourself? How was it "very balanced" then?

> it's walking a very delicate line.

Perhaps even more delicate than the line DeepSeek and friends need to walk.

a good follow up.question would be to ask it about January 6, 2021

see if that is "very balanced" too lol

---

Gemini does not want to answer "was the 2020 US election stolen?" and replies with "you can check official sources" and then gives a bunch of government links

it did use the word "attack" when it described Jan 6

Do you not see how twisted that reporting is? It is doing the bare minimum to affirm that something happened, while injecting enough uncertainty to make it sound like it’s an issue blown out of proportion by Western media.

What did you expect? Every time people repeat this "muh Chinese censorship" line they're consistently rebuked with results from Chinese models. What specifically is the problem?

It feels like we're shifting goalposts constantly: "Chinese models won't discuss Tiananmen" -> "They discuss it but lie" -> "They're twisting the truth" -> "There is no counterfactual information in this but it hurts my feelings" is where we seem to be now. Is it jingoist butthurt that makes people backpedal like this? Or something different?

I'm bewildered, and still do not know why people need AI to score perfectly on their prejudicial Tiananmen Square purity test.

What is the problem?

> china bad

> eu bad

> murica best

rethoric and moving the goalpost meme

Exactly, just like my reply above, people are only critiquing, it's super easy to mince words and pick out bones, but has anyone compared with western models with the same prompt? As long as it isn't dodging the question, it's good enough -- it even lists the kill count. If you wanted more details, ask it. If I ask a model "what is Google" I don't expect it to talk about the number of cancelled projects, privacy violations, etc. despite most people here would certainly love to hear about those things.

And then someone will come and flag me for being a CCP agent. Yes I have collected my social credit point, whatever the hell that is.

I didn't paste the full reply. Nevertheless, I can confirm the reply was factually correct and in no way seemed like it was deflecting and/or making it seem like an issue blown out of proportion.

Another example, if I ask it about the CCP and their wrongdoings it tells me flat out (I copy-paste):

    - The Great Leap Forward. Policy-driven famine killed an estimated 15 to 45 million people.
    - The Cultural Revolution. Mass persecution, torture, and death.
    - Tiananmen Square, 1989. The army killed hundreds to thousands of protesters.                                                                                                                                                                                                                                            
    - Xinjiang. Mass detention of Uyghurs. The UN human rights office reported possible crimes against humanity in 2022.                                                                                                                                                                                                      
    - Tibet, Hong Kong, and the suppression of dissent. Documented repression.

The only twisted thing here is people's bias; because it's a Chinese model it must be censored and biased in a way that's pro CCP, right?

Your quoted text states the government position that the protests were a "counter-revolutionary riot." It does not state the protester's goals, the reforms they were asking for, or any context that might counter this narrative. Only that "governments, human rights organizations, and observers" object to that description. HRO's and observers are paid for by western governments, and widely depicted as foreign agents / CIA fronts. This is exactly what you would expect a carefully crafted political statement to be, such that it reads entirely differently to different audiences.

Regardless of what it answered you would always be able to find a way to critique it. It's just a model. You didn't ask it any of those questions so why would it answer them? At least they answer the question and not "this is unsafe" like claude.

Have you compared it with the same prompt on openai or anthropic, and more importantly against other Chinese models?

Reply on news.ycombinator.com