this post was submitted on 29 Jan 2025
151 points (92.7% liked)
Technology
61227 readers
4363 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Of course it's a national security threat, it's just more proof that the US economy is just a giant ponzi scheme.
If China can do it better on a budget of $6m in 18 months with low end equipment, then why does it take an American company 10 years, half a trillion dollars, and the entire nation's supply of high-end graphics cards?
The model isn’t afaik. I.e., if you download one of the models and run it locally. It’s the app with folks pasting proprietary, company secret, etc data into it.
Really, it’s the same problem as with ChatGPT, but now an organization in another country has your data. I guess we’ll see if our new techno bro overlords try to use this to their advantage across the board to limit competition, even from local processing.
Taking bets.
I just find it amusing how when proprietary data/company secrets/whatever are being sent to openAI it's a matter of "that was irresponsible don't let it happen again" but some guy in Kentucky isn't able to get a detailed description of Tiananmen Square from the US perspective without a little effort and it's the end of national security as we know it.
Same with the tiktok ban. How many classified military secrets do we think some regular dude in a trailer in Alabama really has on his phone?
"National Security" in the US is literally just code for rich people's bank accounts at this point.
The War Thunder forum is a greater threat to "national security" than any of this AI whohash. Something, something, nickle...
How about from the perspective of the pro-democracy protestors who were there? Don't turn a brutal crackdown on people trying to gain some control of their lives and their country as an East vs. West problem.
I'm not here to discuss the validity of Tiananmen Square, that was just the example I keep seeing used.
Why does it matter if one source doesn't provide the official CIA story? You can look up how America views that event anywhere.
How is that censorship any worse than US tech companies blocking you from being able to say the word "Republican" in a negative context?
Also, you left out the most important part "without a little effort." Deepseek will happily tell you anything you want about Tiananmen Square from any perspective you ask it with a little creative prompting.
I only commented because you said something stupid about Tiananmen Square that chapped my hide. The rest of it is fine. I'm only responding again because you doubled-down. The CIA version? I was alive at the time and followed the news, including live TV reporting, for days.
Say what you want about the politics of US News channels at that time, they weren't all in lock step with the CIA. I watched as Zhao Ziyang visited with the hunger-striking students, and I watched as the tanks rolled in. Don't try to revise history because you need the US to be the #1 bad guy. China had a chance to reform, and they cracked down instead.
You read wrong.
K
Show me. You can get it to say some slightly vague things about it, but you can’t have it say “anything from any perspective”. Can’t we just leave out a conspiracy theory while discussing AI?
Like I said, it just takes creative prompting
Hmm, I was talking about local running, but apparently that's just the non-distilled version, and someone got a very straightforward explanation with the 14B distilled version.
Then you should've specified that those were the parameters you wanted. Answers and thought processes will vary based on the prompt provided.
My point is that you can still use creative prompting to get answers you want that should be blocked due to its safety constraints. My point isn't that there's no guidelines to work around.
I'm not an AI researcher nor do I work professionally with AI so I'm not familiar with 100% of the background processes involved with these LLMs but if the question is "can you get Deepseek to talk about Tiananmen Square" then the answer is yes.
Depends. Are there any military things in Alabama?
Ever heard of Huntsville?
Uh . . no . . no we hav- I mean, I haven't. Could you maybe upload some video of this "Huntsville"? Preferably in angles that are not otherwise obtainable via a sweep of existing Internet resources?
We uh have a friend who is very interested in Huntsville, maybe you could focus on some of the city's important areas and uh, industries.
An absolute fuck ton
Well yeah, it’s obviously more of a risk to send directly to your rival than internally. Both are risky but one is much, much worse.
And what exactly is the average person sending to China that's such a threat to US global Imperialism?
Sure, ban it on government devices or whatever you want to do, but why should civilians be punished because the government can't embezzle as efficiently?
A lot of stuff. You analyze that data, you can refine your psyops.
And why should I be more worried about a hypothetical psyop that i might experience than the current psyops that I am experiencing?
You are experiencing psyops from every direction. Some are just more obvious than others.
Am agreeink with these quesiton. Too many US politicians are not great leaders like Putin. And China.
Stupid users send private keys and other secrets to their AIs all the time. This is a big fucking threat to US global imperialism.
The US trusts OpenAI (even if they shouldn’t) to not send hackers after US companies. They definitely don’t trust Chinese companies to have the same restraints.
Unfortunately that's just a danger on the internet. Stupid users are gonna get scammed whether it's a stock trading AI that empties your bank account when you link it or a Nigerian Prince who just needs $5000 so he can unlock his fortune and repay you $100,000.
Even then, what national security upending information does the average citizen have stored on their phone that they're just whimsically uploading anywhere that'll take a PDF? Like I said, I understand restrictions on devices used by government officials for official purposes, but to ban it unilaterally for civilian use as well seems excessive.
"Internally"
This guy really loves his oligarchs and their government haha
Nah, I’m speaking from the perspective of the US, since the article is about US policy. The decision making is obvious when you’re thinking at a national protectionist level.
Obviously privacy violations are bad for the user regardless. Never trust your corporations or government!
Youa re speaking on behalf of daddy Sam?
Damn boy, is u elite?
The hell are you talking about? It’s right there in the article. But maybe you didn’t read it?
Ad hominem attacks like you are using are a sign you don’t have anything useful to say.
No data is sent to servers if you run it locally.
You can download the model.
If you download the app, though, yes thats going to their servers.
China built it in cave, with a box of scraps!
First time you do something is always harder. OpenAI just didn't think it was 1000x harder and thought they'd have more time to cash in.
Myself, I think that being able to throw billions of dollars at hardware, and their focus on next-quarter results discouraged them from putting in the human effort to analyze and optimize their process. It turns out there were some fantastic optimizations to do.
MVP in Technology. OpenAI just sat around throwing salt to the wind piling up "value" until they can convince people it is worth some obscene amount of money to sell out. Once you give someone a literal milestone and show them the path, boom.
This really really feels like a real life Tortoise and the Hare story. Like real hard, and I don't feel the least bit bad for the hare.
oh yeah, not denying that the prototype will be more expensive and resource intensive than following versions, but the whole "US overspends on novel technology, China blows that technology out of the water and shows this tech is both accessible and affordable, US bans Chinese product because American companies don't want to compete" shtick is just getting old
Hmm, what are the previous examples?
Electric cars and Huawei are two recent examples off the top of my head
They spent tens of billions of dollars already. Of course they don't want cheap competition. That's basic capitalism, nothing pro or against China.
Distilling OpenAI and Llama models probably also helped quite a bit
Although I must admit, that the architectural changes are pretty cool
but I have to add, that I've just started reading into the topic a few weeks ago and don't really have any real practical experience, besides checking out some huggingface docs I got linked yesterday and stupid me hasn't thought about looking there...
So everything I say is probably bullshit o:-)
Sure it made the training process faster, but this still takes a fraction of the energy to generate a single output compared to other LLMs like ChatGPT or Llama. Plus it's open source. You can't discredit a technological advancement for building upon previous advancement, especially when doing so with transparency.
As I said, the architectural changes are quite cool
As far as I've understood it mostly comes down to splitting it up into multiple expert systems, so you don't need to activate the complete system with every request
But I've only scratched the surface...
Also, open source... The weights are made publicly available.
None of the training data or systems
Edit: regarding "open source":
Also Meta's Llama is on huggingface, just like deepseek. I still wouldn't talk about transparency here
Deepseek used distillation, which is a way of extracting training information from other models through querying the model. In other words, some of the advances came from examining OpenAI’s models. Being first is hardest and took brute force.
Not that America doesn’t have its own problem. But what do the suicide prevention nets looks like at your office. Because they’re everywhere in china because of shitty working conditions. This is how they do shit so cheap.
Unfortunately China is the future of USA and not vice versa.
I don't understand why you're getting downvoted. Labor laws in China are shit. A ton of people there work way more than 40 hours a week for less money than US Americans get, live on company "campuses", and have suicide nets.
Many people online have been radicalised into thinking they have to be 100% for side A or for side B.
When you put any criticism towards A or B, the supporters go absolutely wild. They will deny any problems with the side they've chosen.