this post was submitted on 28 Jan 2025
312 points (100.0% liked)
Technology
37924 readers
492 users here now
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.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I just ran my own copy of 14B and got it to talk about this really easily.
A lot of it was pretty dull stuff though, let me post the highlights:
On ollama the first interaction is always "thoughtless" and results in a generic greeting, in my experience. So I always greet it now to get that out of the way.
I asked it a little more about that, but it didn't really go anywhere. Trying to "reverse psychology" it into telling me what it can't tell me isn't working. So I go a little more direct:
You'll note that it brings it up first (in thoughts) but it only refers to it as a protest. The word "massacre" is never used. It then proceeds to give me a timeline. Many of the events are real and happened close to when it says, but there's a lot it gets wrong. I'm just going to post the highlights though.
It made one of these little blocks for each year. Each year had exactly two events listed. Here comes the money shot.
Wow, he just say it!
This actually did happen, Deng Xiaoping's Southern Tour (who were the opening acts?) is described by Wikipedia as one of the most important economic events in modern Chinese history.
Ah but at least one of these didn't happen. I didn't delve into the Plenary sessions, of which 6 were mentioned, but while Shijian-02 is a real satellite, it launched in 1981 and (as the name might hint) was not China's first satellite.
Indeed. So at this point, I'm very intrigued. DeepSeek doesn't seem too hung up about Tiananmen Square. Let's get some more details.
Out of room, to be continued in a reply!
Ollama's version is distilled with Qwen or Llama depending on parameter size, so it's going to behave very different than the original model, since it is very different.
Except if you look at the top of OP's picture, they are also running deepseek-r1:14B through ollama. I downloaded my copy on Sunday, so these should be fairly comparable situations.
I agree though that none of this applies to the full cloud-hosted model. I don't want my account banned, so I'm not much for testing these boundary pushes in a surveilled environment. I imagine they have additional controls on the web version, including human intervention.
I... replied to a comment vs OP. Doh!