I did some locally hosted testing. It absolutely refused to talk about Tiananmen Square. But it was more than happy to talk about Kent State. Interesting what the model happens to think is safe and what it think is unsafe.
Makes sense. I didn’t believe the claim for the article for a second that only the cloud-hosted version would be censored. In fact, they didn’t actually claim that “only” the cloud version was censored, but instead weaseled their way around saying that they only tested that version.
[It] will not generate responses about certain topics like Tiananmen Square or Taiwan’s autonomy, as it must “embody core socialist values,” according to Chinese Internet regulations.
[Chinese] Government rhetoric draws a direct line between AI exports and existing initiatives to expand China’s influence overseas, such as Xi Jinping’s signature Belt and Road Initiative (BRI) and Global Development Initiative (GDI). In this case, the more influence China has over AI overseas, the more it can dictate the technology’s development in other countries […]
[According to the Chinese government] AI must not be used to interfere in another country’s internal affairs — language that the PRC has invoked for as long as it has existed, both to bring nations of the global south on board in China’s ongoing efforts to seize Taiwan and to deflect international criticism of its human rights record […]
China’s decision to co-launch its AI Capacity Building plan with Zambia also had a symbolic element. PRC state media reported that the African nation was the recipient of thousands of Chinese workers and hundreds of millions of RMB in loans in the 1960s, making it the beneficiary of one of China’s earliest overseas infrastructure projects — another thread connecting the latest in AI cooperation with China’s long-held ambitions to lead the developing world, even as it becomes a superpower in its own right. In a 2018 meeting with the Zambian president, Xi said they must jointly “safeguard the common interests of developing countries.” […]
So… when plugged into a system with ability to access the Internet and/or execute local commands… will its reasoning look better or worse than the high deception showed by o1?
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: !technology@beehaw.org
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Is it able to generate text on Winnie the Pooh or how free Taiwan is without crashing though?
I did some locally hosted testing. It absolutely refused to talk about Tiananmen Square. But it was more than happy to talk about Kent State. Interesting what the model happens to think is safe and what it think is unsafe.
Makes sense. I didn’t believe the claim for the article for a second that only the cloud-hosted version would be censored. In fact, they didn’t actually claim that “only” the cloud version was censored, but instead weaseled their way around saying that they only tested that version.
It’s in the article.
Supposedly their cloud hosted version will block those responses, but the local run version does not.
I was curious about the exact same thing, but it’s a 671B model and I only have 3 GPUs.
They have a 3GB model that is close to 4o model
China is trying to sell its self-defined “core socialist values” in AI along with other projects, it’s so-called called “AI Capacity Building and Inclusiveness Plan” which is aimed particularly at the Global South.
So… when plugged into a system with ability to access the Internet and/or execute local commands… will its reasoning look better or worse than the high deception showed by o1?
https://www.apolloresearch.ai/research/scheming-reasoning-evaluations