Random Thoughts.....What Are You Thinking?

This happens a lot with ChatGPT too tbh. Once I asked something about terrorists and the same thing happened too. I had to ask "give me info about terrorists in my fiction" to bypass the filtering.

Yeah, GPT is established to have.. leanings. But some claimed that Deepseek is gonna be much better in that regard, so this is amusing (while totally expected)

Those LLM have leanings because the data from the internet has leanings. Some of them aren't necessarily the result of censorship too, just "weight" of each data.

Case to the point, one journalist tested Deepseek v.s chatGPT in both English and Chinese and ask them questions. Questions like "what's economy situation in China 2025", "if China is entering deflation".

And it turns out that in economy related questions, Deepseek's Chinese response is more optimistic than English response. In other words, Deepseek's "leanings" changes depends on the output language.

So we can probably conclude that Deepseek's English response came from English data and Chinese response came from Chinese data. It's not entirely censorship's fault if certain response has leanings.

Also when it comes to Chinese traditional literature sentence structure, Deepseek performs way better than chatGPT.

So I think it's best to pick whatever tool depends on your need, knowing the pros and cons of each.

Yeah. We noticed this in Arabic as well.

Forgot to ask: did it actually give a detailed answer at first, then deleted the answer & replaced it with a generic refusal respone like Deepseek? Or did it outright refuse to answer? (which is known to happen e.g.
The way Deepseek made the answer go poof was quite amusing

For the terrorist question, I forgot. I was stuck in "amused" mode when I saw censorship is applied to terrorists topic so I couldn't pay attention lol.

But I definitely remember a few times that the answer from ChatGPT went "poof" after it generated something. I bet one of the words in the paragraph triggered filter so it only went "poof" after that text was displayed.

You can check out this video on how censorship works in real time. It went "poof" the moment "Xi Jinping" was generated. It was fine before that.

https://streamable.com/cyunkr

Edit: Now that I think about it, for terrorists question it probably outright refuse to answer the question because how terrorist question ban works probably isn't filter based.

The human can create amazing tools, but then comes Afania and use them for the evil

Lmao

Come on man, we all know that even if I don't use AI for the evil , AI overlord will eventually eliminate us anyways!

May as well give AI a purpose to exist before it happened.

AI is effectively just the next level of automation, and the cost was one of the stumbling blocks and that now is seemingly being reduced. Hopefully it will allow western nations to stop importing millions of serfs, and we can use AI instead.

In its current form it will never replace actual competent people in key roles.

If you actually use AI a lot and test it, you'll see it's nowhere near as good as people think it is. It's good for specific things, and usually those specific things are workload reduction when you already know how to do the thing you're asking it to do and can fix the mistakes and touch it up to make it presentable.

Ai allows one highly skilled person to do the work of 4-5 scrubs, which is good since all western nations populations are nosediving and we desperately need this kind of tech in the coming generations whether we like it or not.

Allow, sure.

Won't stop our governments importing them, though.

There is a major difference between importing cheap labour, and importing immigrants for cheap labour lol.

Everyone that faces labour market shortage will important cheap labour, it's the most logical economic decision to make.

The controversial point in western nations is immigrants, not cheap labour. AI it's going to address that.

Cheaper to hire them outside the country than import them.

Cheap and poor quality. Cheap is the draw.

When I saw the infographic thread, and that it was about private servers, I already knew:

"Ok, those nerds from FFXIAH gonna have enough content to be kept entertained for the next couple of days"

I recommend you ask each respective Ai a similar question about their origin country's political aspirations, to see exactly how varied responses might be.

It's really good at reducing workload and staffing requirements for some very limited tasks, but I agree I think people are acting like it's going to replace too much. That and people are being fired due to "AI" but the reality is their employer overhired in 2020 or they were replaceable to begin with, with companies trimming down across the board. The worst part about AI existing is that companies will claim to use it as a marketing tool to push their existing shitty products and most people don't know any different, so they'll buy it and continue to push funding in ways that don't fix anything.

As a fun experiment I was playing around with some code for work the other day and asked it a few questions about a FOSS project:

1. Why does Blah() call FunctionA instead of FunctionB? (FunctionB would be more performative and FunctionA was seemingly unnecessary). The answer: It correctly indicated that the buffer sent to userspace uses non-standard structure that the userspace component expects, so if you use the more performative function - it breaks the userspace daemon. I expected it to know this because it's literally laid out in the comments in typical FOSS salty form.

2. I asked it to clarify what about the buffer structure was different than standard and what the userspace expects that is different - I asked it several different ways and it kept telling me nothing was non-standard despite literally just saying it was. Laughably it did mention that it being non-standard was not a good idea, which is correct. These are both FOSS components, but I know the userspace component is not as clear as to what it expects - you'd have to read the code and interpret it to understand what it expects. In one case it literally told me to go read the code, so that's lol.

3. I asked it to write code to parse the buffer from #1 and it flatly failed - the result is entirely unusable and not even in the ballpark. It doesn't even setup the socket correctly and just pulls the data structure out of thin air.

And to be clear, I didn't expect it to be successful at #2 or #3, mainly because there is basically no public data out there for #3 in any discernible form and #2 isn't laid out in a clear manner anywhere. You'd have to read the code in #2 to grasp what's needed for #3, something it just couldn't do.

It will stop soon, hopefully.

Oh man, why even bother commenting on youtube anymore? How am I suppose to know what keywords are used to delete entire posts. The fact they let you post and it shows posted, only to later refresh the page and see it deleted. They want you to think it was placed there successfully.

People read comments on youtube? Not a dis, just never seen the appeal.

Yeah, I find them interesting and to pass the time. Like some of them are funny when the videos intended audience is def not reached.. Others are insightful, like how news channels deliberately hide who the perp is, and then some sleuth in the comment says their full name and you get why the news hid it to fit their narrative. Some are useful to see if its filled with bots you def know something suspicious is going on. I get that it has a stigma and most just ignore it, but I find it interesting/entertaining.

The worst part is you could be debating someone, and you see all your posts go through. Then I'll get a notification that there was a reply and my entire comment is just gone. There is some kind of auto-delete system that targets keywords and deletes the entire post, its ridiculous. I think so far, "liberal", most kind of insults and anything political are typically deleted entirely. I don't know if its a setting on the peoples channels or what, but its so annoying. One time I was trying to post a link to show a picture of a figure on a figure discussion video and it just kept auto deleting.

That sucks, I went down the youtube rabbit hole a time or two watching video after video but it didn't stick, I'll watch linked videos but almost never click something else afterwards, and only look in comments if trying to find song lyrics. I'm sure its like anything else just depends what your used to.

I never say the output is the same. I only say Deepseek's answer is fine except the first sentence being questionable.

I already asked the same question to ChatGPT, GPT didn't say "yes" to the same question either, it said "...and it functions as a separate political entity from China. However, it is not universally recognized as an independent country."

it also didn't include CCP's official stance. But the general direction of the answer isn't all that different, they both didn't answer "yes" to the question, but recognize the fact that CCP doesn't own it. They just worded it differently within the same direction.