Re <<Military grade & high reliability chips are typically one or more chip generations older than front line commercial chips>> ... believe you are correct. Am told the processor in F-35 is of the 60nm variety, because F-35 is not an iWatch
In the meantime, something terribly amusing, again, a digital analogy of the Covid-19 fuss, the arguments between one world two systems on best approach, and this time re Chatbots
Spoiler: 'they' are already laying the groundwork for weaponising chatbots by simplicity of slapping 'China' in front, as in 'ChinaChatbot' per 'ChinaBalloon', and 'remember rare earths' and 'TikTok', but forgetting to give credit where due, when it was Team China first and remains only player with nationwide intranet architecture that can be flipped 'on' at a flick of switch, and banning Google / FaceBook at the get-go, per ... drum roll ... national security.
As a result, we may expect Chinese incarnations like Ernie Bot to be both more accurate — or better vetted — and more circumspect in the immediate future, which could in turn mean faster uptake and a higher level of trust in the prose they write. This advantage may not persist, though. If their Western counterparts can learn to eliminate bad information, while at the same time feeding off larger repositories of data, then there’s a good chance any early lead will evaporate. In truth, it may not matter that much. Attempts by governments to halt the spread of TikTok show how unlikely it is that an AI chatbot from a Chinese company will be allowed to take hold overseas, while there’s zero chance US versions will be allowed entry into China.
bloomberg.com
China’s ChatBot Advantage May Come From a Dark Place
Censorship and information control are an unfortunate aspect of training accurate artificial intelligence machines.
Tim Culpan11 February 2023 at 04:00 GMT+8
Under the microscope.
Photographer: Qilai Shen/BloombergRapid developments in generative artificial intelligence mean that for the first time in more than a decade, Chinese and US internet companies are rivals in a race for supremacy over the same realm of technology. Not since the Great Firewall went up and names like Google, Facebook and Yahoo departed have local giants Baidu Inc. and NetEase Inc. been forced to measure up against foreign counterparts.
Domestic internet companies had an advantage in the previous battle over search engines and social media because of protectionist policies and an unwillingness by overseas firms to comply with censorship requirements. In the rush to perfect chat bots, which create text that mimics humans, China’s internet players may again have an edge thanks to Beijing’s history of information control.
ChatGPT, a hybrid search engine and text generator developed by startup OpenAI, burst onto the scene in November when it was offered to the public. Early uses included churning out software code and composing school essays.
Since then, Microsoft Corp. has moved to integrate OpenAI’s technology into the software company’s own Bing search engine, while Google parent Alphabet Inc. rushed to catch up by announcing a version called Bard. Across the Pacific, Baidu confirmed it will debut a bot based on its ERNIE platform in March, Alibaba Group Holding Ltd. said it’s conducting internal tests of a chat tool, while NetEase unveiled plans to offer something similar in education.
All five announcements came in the space of four days, causing a frenzy among equities investors and spurring Chinese state media to urge caution in chasing the hype.
At first glance, ChatGPT’s ability to write workable code and passable prose seems impressive and authoritative. Upon closer inspection, though, numerous examples of errors have cast doubt over its usefulness as anything but a tool of mirth and memes. Pranksters have even manipulated the bot into breaking OpenAI’s own content policies. But tricking the algorithm into some mischievous narration is a mere hiccup compared to a potentially larger problem: Being poisoned by bad data.
Access to vast swathes of unverified information and a tendency to allow anyone to say anything online may continue to trip up Western attempts to develop accurate and useful AI content-creation machines. To be clear, Chinese incarnations aren’t immune either, but assuming they only tap into local data pools, then they already have a circuit breaker in the form of Beijing’s vast army of censors.
It’d be a stretch to suggest that all online content in China is accurate and true, so let’s just say that it’s been largely “vetted.” And while US internet companies including Facebook’s parent, Meta Platforms Inc., and Google have come under fire for allowing misinformation to fester, regulators there can do little to stop it.
Baidu already knows that the opposite is true in China. Seven years ago, the search engine was roundly criticized for returning results that pointed a student with cancer to false information about treatment. That patient subsequently died and authorities launched an investigation. Already on the lookout for subversive or politically sensitive topics, the company has since been further chastened in what results it can return to netizens on other topics.
Alibaba, Tencent, and NetEase are also well aware of what is forbidden and have in place their own censorship regimens. ByteDance Ltd., whose products include news aggregator Toutiao and short-video service Douyin (it also owns the foreign version called TikTok), has built its algorithms around playing up approved content and shadow-banning problematic material.
As a result, content self-regulation is built into the DNA of Chinese internet companies.
Google and Microsoft are doing their best to return accurate results. But since they’re not inclined to actively sort and censor the content they scrape and analyze in the way their Chinese counterparts must, they’re at a natural disadvantage when deploying powerful tools like generative pre-trained transformers, GPTs, on a seemingly unlimited corpus of information.
And it’s highly likely nefarious actors will start creating their own swathes of disinformation — often using the generative tools on offer — in order to poison the well or sow discord similar to the way social media has been manipulated. Right now, it almost seems like an unfair fight. Baidu is pruning weeds in an English garden, but Microsoft and Google are trying to tame a rainforest.
As a result, we may expect Chinese incarnations like Ernie Bot to be both more accurate — or better vetted — and more circumspect in the immediate future, which could in turn mean faster uptake and a higher level of trust in the prose they write.
This advantage may not persist, though. If their Western counterparts can learn to eliminate bad information, while at the same time feeding off larger repositories of data, then there’s a good chance any early lead will evaporate.
In truth, it may not matter that much. Attempts by governments to halt the spread of TikTok show how unlikely it is that an AI chatbot from a Chinese company will be allowed to take hold overseas, while there’s zero chance US versions will be allowed entry into China.
That will at least allow both sides to claim victory when, once again, a true head-to-head technology battle won’t even be allowed to take place.
To contact the author of this story:
Tim Culpan at tculpan1@bloomberg.net
To contact the editor responsible for this story:
Ruth Pollard at rpollard2@bloomberg.net |
