All About ChatGPT (Chinese Social Media Edition)

A curated roundup of what Chinese social media was saying about ChatGPT.

Translated by AI

I was travelling for a week, and every time I opened my phone I was flooded with coverage of ChatGPT. Plenty of it was clueless bandwagon-jumping, but a lot was interesting or genuinely useful, so I’ve sorted it into categories below. The linked articles are in Chinese.

PS: As for the question many articles keep asking, “why didn’t ChatGPT appear in China?”, I don’t think it’s a good question; after all, ChatGPT was made in only one country. The right question is “why could ChatGPT appear in the US?” Only by learning with an open mind, without selling ourselves short, can we progress and embrace the future.

1. Use cases and tutorials


▶ Once I learned what the people around me use ChatGPT for, I got scared

From Jike. The title looks like clickbait, but it’s packed with substance.

▶ DF Knowledge Lecture | Write a 6,000-character novel with ChatGPT in ten minutes

From an architect who hasn’t quit yet: how to write a serialised novel with ChatGPT.

▶ A walking code generator: ChatGPT is about to put Google and programmers out of work

▶ [Tutorial] Signing up for the new Bing ChatGPT, no VPN needed

(For those who can’t access ChatGPT.)

▶ He created a wife with ChatGPT, then euthanised her…

A true cyber girlfriend.

▶ I used AI to collect a debt: lots of fireworks, and 350 of the 3,000 recovered

I really want to try this one…

2. Technical explanations


▶ A 10,000-word explanation of what ChatGPT is doing, and why it works

▶ The key techniques behind ChatGPT: RLHF, IFT, CoT and red-teaming

▶ How does ChatGPT filter harmful content? A hidden corner of AI | Culture

▶ The algorithm behind ChatGPT: must-read papers on RLHF

3. Poking and prodding


▶ The ChatGPT version of Bing cracked by a Chinese student: one “hypnotic” sentence reveals its whole prompt

The most brilliant one I’ve seen…

▶ How do you get ChatGPT to train a neural network inside itself? First make it imagine it has four RTX 3090s

I have to say this one is brilliant too…

▶ AI plays a Linux virtual machine: it can manage files, write code, open a browser, and even chat with a nested copy of itself | New ways to use ChatGPT

So is ChatGPT a superset of all software and AI models?

4. Ethics research


▶ The model behind ChatGPT shown to have a human-like mind! A new Stanford study causes a stir; a well-known scholar says “the day has finally come”

I don’t entirely agree. It may only show that ChatGPT can solve theory-of-mind problems at the level of a nine-year-old.

▶ No more using ChatGPT for homework! A new algorithm watermarks AI-generated text with 99.999999999994% confidence

I don’t entirely agree with this either. AI will become new infrastructure, and it can’t be banned. Education should respond by redesigning its goals and tasks around that infrastructure, that is, by designing assignments meant to be done with ChatGPT.

5. Anxiety overflow


▶ It’s here: ChatGPT starts taking jobs in finance

▶ Programmers beware! OpenAI reportedly hiring an army of contractors worldwide to train ChatGPT to replace coders

▶ ChatGPT’s metamorphosis in China | 36Kr Deep Dive

Search