• Biblical Families is not a dating website. It is a forum to discuss issues relating to marriage and the Bible, and to offer guidance and support, not to find a wife. Click here for more information.

DeepSeek AI

The U.S. Navy has instructed service members to avoid using the Chinese AI platform DeepSeek, citing "potential security and ethical concerns," according to CNBC. An email sent to "shipmates" in recent days, confirmed by CNBC on Tuesday, referenced the Navy's AI policy and emphasized the importance of refraining from using DeepSeek. The memo warned service members against using the platform "for any work-related tasks or personal use" and instructed them to "avoid downloading, installing, or using the DeepSeek model in any capacity."

U.S. Navy Bans DeepSeek Over 'Security Concerns' As 'Substantial' Evidence Emerges Chinese AI Ripped Off ChatGPT | ZeroHedge

ZeroHedge - On a long enough timeline, the survival rate for everyone drops to zero
www.zerohedge.com www.zerohedge.com
 
MemeFan said:
What US strategy is/was can be best described as max chips. With (reportedly) 50X better efficiency DeepSeek wouldn't need all GPU farm.
Click to expand...
How did they achieve this 50x efficiency though? That’s a huge disparity and there’s no good explanation for how this miracle was wrought.
 
The Revolting Man said:
How did they achieve this 50x efficiency though? That’s a huge disparity and there’s no good explanation for how this miracle was wrought.
Click to expand...
Yes, there are very good explanations all over the place, as they published detailed papers explaining what they did, and these are summarised by techies around the internet.

As far as I understand in my very layman reading of it, they made the training more efficient by sacrificing precision in areas that don't actually matter much, in order to prioritise training just in those areas that actually mattered the most - in other words, they didn't waste massive amounts of computer resources training it to do things it wouldn't ever do much. And they simplified the model by breaking it into a "committee of experts", with multiple small sub-models that are trained just on a portion of human knowledge and focus on just doing that one thing well, so when you ask the model for something it doesn't have to run the entire thing but can just run the smaller model that is an expert in that field. But if you want the technical details there are articles all over the place, can find the links if you like.

Yes, they do appear to have copied a lot from ChatGPT - but they've released theirs open source, so every other model will now copy from them also. This is how science advances, each person taking all available knowledge and then adding one further step.
 
FollowingHim said:
Yes, there are very good explanations all over the place, as they published detailed papers explaining what they did, and these are summarised by techies around the internet.

As far as I understand in my very layman reading of it, they made the training more efficient by sacrificing precision in areas that don't actually matter much, in order to prioritise training just in those areas that actually mattered the most - in other words, they didn't waste massive amounts of computer resources training it to do things it wouldn't ever do much. And they simplified the model by breaking it into a "committee of experts", with multiple small sub-models that are trained just on a portion of human knowledge and focus on just doing that one thing well, so when you ask the model for something it doesn't have to run the entire thing but can just run the smaller model that is an expert in that field. But if you want the technical details there are articles all over the place, can find the links if you like.

Yes, they do appear to have copied a lot from ChatGPT - but they've released theirs open source, so every other model will now copy from them also. This is how science advances, each person taking all available knowledge and then adding one further step.
Click to expand...
We’ll see. I’m skeptical of the whole AI hype anyway and of course despise communists with every fiber of my being. Creating a 50 times more efficient model will have been an impressive feat if it’s true.
 
The threat remains a faster, more efficient, More Brilliant Psychopath.

PS> And they WILL worship that 'god of stone' (in this case, silicon) too.
 
The Revolting Man said:
It’s starting to sound like the whole Deep Seek thing may be a lie. The company that allegedly built it probably has billions of dollars worth of high end American chips. It may just be a Chinese copy of what America already has.
Click to expand...

Even if it is a copy, you can run it on an rPi, mind you without an external gpu it only does a token and change per second, but...
 
The Revolting Man said:
How did they achieve this 50x efficiency though?
Click to expand...

They probably didn't. Remember this is coming from China and they routinely make claims that are utter nonsense. If the CCP wants everyone to say 2+2= Cat then that's what everyone will say and they'll all act like they believe it too.
 
Mark C said:
Again, the back doors allegedly (and I don't doubt it in the least) access the BIOS, and thus the hardware.

And a true AGI, of the Terminator "Skynet" level, would cut through those 'safeguards' like a knife through hot butter.
Click to expand...
Virtual machine can't access HW. And any AI would have to pass triple protection. In order: operating system, hypervisor and microprosser itself.

All defended by best bug catching tehnologies ever in existence with most critical thinks defended by formal validation which proves unbreakability.
 
BiblicalLiteralist said:
for anyone that wants to know how AI model training work listen to this.

Click to expand...
Even without watching it looks like making something made to scare people. Nooooo, climate change isn't scary enough. Nope, here is something scarier which only World Government can stop.

There won't be humans vs Terminators. If it comes to that, it will be one nations Terminators with humans against other nation's Terminators with humans.

New tech rarely replaces old tech*. Usually both are added to humanity tech "skills"

Kevin Kelly once ran award for finder to tech totally not used anymore. Best runner, but not winner was guy with idea of medieval torture tech. Yup, we still use those somewhere.
 
It's worth noting that even if DeepSeek is completely dishonest - they spent more on training than they claim, they stole much of it from OpenAI, or whatever - that in itself is a major upset to the whole tech sector. That means that top end Western tech can be stolen/replicated relatively easily. That still means the West will struggle to stay ahead of the East in this sector - it's just as bad news for markets as if China did this completely independently. If not worse news. Whatever angle you look at it from this is a very disruptive upset to Trump's plans to have the US lead the world in this new field.
 
Here are some additional resources explaining DeepSeek:

stratechery.com

DeepSeek FAQ

DeepSeek has completely upended people’s expectations for AI and competition with China. What is it, and why does it matter?
stratechery.com stratechery.com

www.tomshardware.com

DeepSeek's AI breakthrough bypasses industry-standard CUDA for some functions, uses Nvidia's assembly-like PTX programming instead

Dramatic optimizations do not come easy.
www.tomshardware.com www.tomshardware.com

www.tomshardware.com

Chinese-made DeepSeek AI model records extensive online user data, stores it in China-based servers

What doesn’t DeepSeek collect?
www.tomshardware.com www.tomshardware.com
 
MemeFan said:
Even without watching it looks like making something made to scare people. Nooooo, climate change isn't scary enough. Nope, here is something scarier which only World Government can stop.
Click to expand...
Yeah, more fear = more opportunities for control.

Interesting, I did see that Tesla now has permission for driverless FSD cars to be on the roads on Texas in June this year and other states after that. AI is hitting the roads in the USA. And Optimus is going into production for release next year.
 
The Revolting Man said:
Are you arguing with yourself now?
Click to expand...
No, you blithering idiot. If you could, or would, just READ the post, instead of looking for any opportunity to start a fight and stoke your own hubris, you'd see just WHAT I was disagreeing with.

I bet everyone else here did.

PS> Even if you're too lazy to actually read the thread, you could still just 'click the arrows' to see why you are again so full of it.
 
You must log in or register to reply here.
Back
Top