China creates LLM trained to discuss Xi Jinping's philosophies
(2024/05/23)
- Reference: 1716438734
- News link: https://www.theregister.co.uk/2024/05/23/china_xi_jinping_chatbot/
- Source link:
China's Cyberspace Research Institute has revealed it's created a large language model and conversational AI based on the philosophies of President Xi Jinping.
The XiBot isn't online, at least not yet – according to a Chinese government [1]announcement it's being tested internally.
The service is built on a corpus of government documents to allow citizens to query them using a generative AI interface. Among the docs in the corpus is a knowledge based on the book "Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era" – the president’s definitive political philosophy.
[2]
Books with the same title is widely available in China. The Register has seen it for sale in airports likely to be frequented by Chinese tourists, or members of the Chinese diaspora, translated into local languages.
[3]
[4]
Xi's thought is also omnipresent on Chinese government websites, and news outlets. When Xi delivers a speech of some importance, his words are reproduced online.
Creating Chat Xi-PT is therefore not out of the ordinary because Beijing goes out of its way to ensure Xi's thoughts are known at home and abroad. Bringing his words to a new medium - chatbots - is an obvious extension of existing practices.
[5]
As The Register has often mentioned, generative AI can make things up and confuse facts. In China, however, Xi is always right – making AI hallucination perhaps less of a political problem than it would be in other countries.
[6]Winnie the Pooh slasher flick mysteriously cancelled in Hong Kong
[7]China upgrades Great Firewall to defeat censor-beating TLS tools
[8]Chinese AI censors live-streamed Alpacas – beasts with a very NSFW and political back story
[9]And it begins. OpenAI mulls NSFW AI model output
A wag in El Reg 's offices suggested the next dictatorial chatbot off the production line could be Kim-Jong AI, to immortalize North Korea's dictator for life.
But in the real world, researchers have already imagined similar facilities for your loved ones, or historical figures.
In late 2023, your correspondent met Dr Simon See, global head of Nvidia's AI Technology Center, which collaborates with academic researchers to explore the tech. At the SIGGPRAH Asia conference he revealed he is aware of people who have created AI-powered avatars of their loved ones, based on recordings and images.
He also mused that Nvidia hoped to work with researchers on generative-AI powered replicas of prominent people from history.
[10]
"Could we recreate Roman times? Napoleon?" he asked.
He thinks the answer is yes, because "We have their letters and we know their history."
Your correspondent pointed out that video footage means contemporary historical figures are far better documented than the likes of Napoleon, and Dr See agreed that the potential to revive them as conversational AIs is therefore greater.
I discussed this idea around the dinner table at home, which prompted my family to declare it daft – on grounds that some would welcome recreations of history's worst tyrants, and that freshly voiced expressions of their vile ideologies would inevitably spread in the fever swamps of social media before bubbling into mainstream discourse.
That won't happen in China, where even oblique puns perceived as criticism of Xi are ruthlessly removed from the internet. Any unauthorized AI would never make it online – or if it did, would invite swift shutdowns and unpleasant punishments. ®
Get our [11]Tech Resources
[1] https://www.tjcac.gov.cn/wxdt/qg/202405/t20240521_6630973.html
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2023/03/22/pooh_slasher_flick_cancelled_china/
[7] https://www.theregister.com/2022/10/06/great_firewall_of_china_upgrades/
[8] https://www.theregister.com/2021/05/14/chinese_ai_censoring_some_livestreamed/
[9] https://www.theregister.com/2024/05/09/openai_model_nsfw/
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[11] https://whitepapers.theregister.com/
The XiBot isn't online, at least not yet – according to a Chinese government [1]announcement it's being tested internally.
The service is built on a corpus of government documents to allow citizens to query them using a generative AI interface. Among the docs in the corpus is a knowledge based on the book "Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era" – the president’s definitive political philosophy.
[2]
Books with the same title is widely available in China. The Register has seen it for sale in airports likely to be frequented by Chinese tourists, or members of the Chinese diaspora, translated into local languages.
[3]
[4]
Xi's thought is also omnipresent on Chinese government websites, and news outlets. When Xi delivers a speech of some importance, his words are reproduced online.
Creating Chat Xi-PT is therefore not out of the ordinary because Beijing goes out of its way to ensure Xi's thoughts are known at home and abroad. Bringing his words to a new medium - chatbots - is an obvious extension of existing practices.
[5]
As The Register has often mentioned, generative AI can make things up and confuse facts. In China, however, Xi is always right – making AI hallucination perhaps less of a political problem than it would be in other countries.
[6]Winnie the Pooh slasher flick mysteriously cancelled in Hong Kong
[7]China upgrades Great Firewall to defeat censor-beating TLS tools
[8]Chinese AI censors live-streamed Alpacas – beasts with a very NSFW and political back story
[9]And it begins. OpenAI mulls NSFW AI model output
A wag in El Reg 's offices suggested the next dictatorial chatbot off the production line could be Kim-Jong AI, to immortalize North Korea's dictator for life.
But in the real world, researchers have already imagined similar facilities for your loved ones, or historical figures.
In late 2023, your correspondent met Dr Simon See, global head of Nvidia's AI Technology Center, which collaborates with academic researchers to explore the tech. At the SIGGPRAH Asia conference he revealed he is aware of people who have created AI-powered avatars of their loved ones, based on recordings and images.
He also mused that Nvidia hoped to work with researchers on generative-AI powered replicas of prominent people from history.
[10]
"Could we recreate Roman times? Napoleon?" he asked.
He thinks the answer is yes, because "We have their letters and we know their history."
Your correspondent pointed out that video footage means contemporary historical figures are far better documented than the likes of Napoleon, and Dr See agreed that the potential to revive them as conversational AIs is therefore greater.
I discussed this idea around the dinner table at home, which prompted my family to declare it daft – on grounds that some would welcome recreations of history's worst tyrants, and that freshly voiced expressions of their vile ideologies would inevitably spread in the fever swamps of social media before bubbling into mainstream discourse.
That won't happen in China, where even oblique puns perceived as criticism of Xi are ruthlessly removed from the internet. Any unauthorized AI would never make it online – or if it did, would invite swift shutdowns and unpleasant punishments. ®
Get our [11]Tech Resources
[1] https://www.tjcac.gov.cn/wxdt/qg/202405/t20240521_6630973.html
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2023/03/22/pooh_slasher_flick_cancelled_china/
[7] https://www.theregister.com/2022/10/06/great_firewall_of_china_upgrades/
[8] https://www.theregister.com/2021/05/14/chinese_ai_censoring_some_livestreamed/
[9] https://www.theregister.com/2024/05/09/openai_model_nsfw/
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Zk8T5fwvrdy96bmmZqm89QAAAQI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[11] https://whitepapers.theregister.com/
Paul Herber
'rebuilding a digital Napoleon'
make it out of ARM-in-waistcoat chips?
Can't wait for this to be jailbroken
veti
Props to the first person who can get Xibot to say "Chiang Kai-shek was right all along".
Tubz
President Pooh's 3 philosophies,
1. I am right, the universe is wrong,
2. Do as I say, not as I do.
3. If in doubt, see 1 and 2
Yeah, this is too dangerous for them.
There's no way they're going to let this loose for everyone (even just in China) to poke at. They're going to need censors scrubbing the output at all times lest the Xi-bot say anything (gasp) non-Xi Jinping Thought approved. And all LLMs so far are very easy to jailbreak - there's no reason why China's would be any better, especially with most of it being 'acquired' technology. So anyone who knows what they're doing with an LLM could make Xi-bot say the funniest, most treasonous things (and the bar is very very low for treason).
Probably their best use for it is to auto-generate his usual rambling, vacuous, pronouncements on every subject under the sun for publishing, saving the real Xi a lot of work. Though he obviously loves to pontificate, so you'd have to strike a safe balance between freeing up his time and taking away his joy of preaching.
I can see other cases where something similar could work better, though. A Trump-bot would likely be more coherent than the real thing, would lie - okay, about the same amount. But would definitely be much less malicious, and at least wouldn't reek of filled diapers. And a Johnson-bot would be pretty much indistinguishable from the real thing for speaking, at least? And wouldn't force your liquor cabinet open and guzzle all the good plonk.