ChatGPT-4: threat or boon to the Great Firewall?

March 21, 2023 @ 7:57 pm · Filed by under Artificial intelligence, Censorship, Computational linguistics

« previous post | next post »

"The practical value of LLMs is high enough that it will induce Chinese to seek out the best systems, and they will not be censored by China.”

"Yes, the Chinese Great Firewall will be collapsing"

by  Tyler Cowen Marginal Revolution (March 21, 2023)

Something that the PRC censors had not predicted:

As framed from China:

Fang Bingxing, considered the father of China’s Great Firewall, has raised concerns over GPT-4, warning that it could lead to an “information cocoon” as the generative artificial intelligence (AI) service can provide answers to everything.

Fang said the rise of generative AI tools like ChatGPT, developed by Microsoft-backed OpenAI and now released as the more powerful ChatGPT-4 version, pose a big challenge to governments around the world, according to an interview published on Thursday by Red Star News, a media affiliate to state-backed Chengdu Economic Daily.

“People’s perspectives can be manipulated as they seek all kinds of answers from AI,” he was quoted as saying.

Fang, a computer scientist and former government official, is widely considered the chief designer of China’s notorious internet censorship and surveillance system. He played a key role in creating and developing the Great Firewall, a sophisticated system of internet filters and blocks that allows the Chinese government to control what its citizens can access online.

I would put it differently, but I think he understands the point correctly.  Here is more from SCMP, via D.  The practical value of LLMs is high enough that it will induce Chinese to seek out the best systems, and they will not be censored by China.  (Oddly, some of us might be seeking out the Chinese LLM too!)  Furthermore, once good LLMs can be trained on a single GPU and held on a phone…

Solve for the political equilibrium.

The implications of Chat GPT-4 are only beginning to be felt.  The closest thing we've experienced to lack of bias on a grand scale?

 

Selected readings

 

Censorship

The LL literature on censorship in the PRC is vast, so I shall not attempt to resume it here.  One of the latest installments is this, which comes with a far from exhaustive list of previous posts:

[h.t. Bill Benzon]

March 21, 2023 @ 7:57 pm · Filed by under Artificial intelligence, Censorship, Computational linguistics


4 Comments

  1. Jonathan Smith said,

    March 21, 2023 @ 8:25 pm

    "(Oddly, some of us might be seeking out the Chinese LLM too!)"
    ChatGPT already is a powerful Chinese LLM; tons of example prompts floating around

    And what is the "practical value of LLMs"?
    * write say romance incl. danmei stories, real estate ads, college acceptance letters, and other pablum crucially including
    * the zombie excuse me the real internet, a purpose to which it is doubtless already being turned
    etc.
    so it depends how you define / who defines "value"? ChatGPT in its current form is not an information-processing tool

  2. yakoumis said,

    March 22, 2023 @ 12:51 am

    "Lack of bias", when there has been extensive Reinforcement Learning from Human Feedback (RLHF), is hard to swallow.

  3. Dan DB said,

    March 22, 2023 @ 3:41 am

    What do you mean by 'lack of bias'? ChatGPT inherits the bias of its training data, reinforced by the biases of its prompt engineers. Passing bias/opinion/inaccuracy through an algorithm doesn't remove the bias, it just launders it.

  4. Victor Mair said,

    March 22, 2023 @ 6:26 am

    neutralization

    as in ACID + ALKALINE

RSS feed for comments on this post