What do we know about DeepSeek?

Topic

New Reply

AI By Michael A. Covington On January 27, the Chinese AI company DeepSeek caused so much panic in American industry that NVIDIA stock dropped 17% in o
[See the full post at: What do we know about DeepSeek?]

11 users thanked author for this post.

WSGeo, fisher75, ibe98765, windbg, SueW, Henry, DLivesInTexas, IBM1130, rc primak, lmacri, Slowpoke47

Reply | Quote

Replies

Thanks for this eye-opener.  What is the takeaway for us lowly computer users?

Reply | Quote

That generative AI technology is changing fast, but it takes time to see what each advance actually amounts to.

And that it’s good not to assume that progress will be constant in one direction with no changes.

2 users thanked author for this post.

SueW, rc primak

Reply | Quote

I am a subscriber to NewsGuard, a company that assesses the factual reliability of websites and other sources of information. They have become a globally recognized source of detecting misinformation and disinformation. They recently performed an audit of DeepSeek against its Western competitors. The results were startling and clearly showed that DeepSeek is a tool of the Chinese Communist Party. If you’re interested in seeing what DeepSeek returns when it’s put to the test, here’s the link to the NewsGuard article:

https://www.newsguardrealitycheck.com/p/deepseek-ai-chatbot-china-russia-iran-disinformation

The link to NewsGuard, the company, is:

https://www.newsguardtech.com/

1 user thanked author for this post.

Andy M

Reply | Quote

If Deepseek’s code is truly open source, then it can be examined by anyone with the time and the technical skills to understand what it’s doing. And if there’s bias or filtering, the filters can be revised to comply with standards of free countries or open source communities. But at the end of the day, this LLM has upended the industry, making it obvious that there are fast, efficient and low-cost ways to build, train and deploy LLMs. This residue is not lost on open source developers.

I would caution anyone curious about Deepseek not to use their apps. But that means hosting your own version, which would still have bias and censorship baked in.

-- rc primak

2 users thanked author for this post.

ibe98765, Slowpoke47

Reply | Quote

I think we can assume China is for China.  Chinese leadership has confirmed right along that they see China as the rightful world leader and will take whatever steps it deems necessary to make that a reality.

China is currently building a ginormous underground military complex that would survive a nuclear hit.  When completed, in about 2 years, one could conclude China might be more willing to go to war with the US over, for example, Taiwan.  During that 2-year window, they may succeed in their as yet failed efforts to develop a missile capable of delivering a nuke to North America.

And a tactical advantage for China is that Xi, as the holder of supreme authority, needs no approvals to substitute action for words.  Looks like the only thing holding him back is lack of sufficient weaponry.

 

 

 

1 user thanked author for this post.

Andy M

Reply | Quote

Since crypto has moved on from the days of needing vast amounts of hardware and energy to mine, I see no reason why AI won’t do the same or why anyone was surprised. Ai is new and it will task a vast effort to get that ball rolling, but like most balls, once rolling downhill it’ll pick up speed.

1 user thanked author for this post.

rc primak

Reply | Quote

Crypto is only valuable when you have to pay for something.

However, in our future, when machines do all the work and AI’s are in charge, everything will be provided for free.  Humans will not need to work.  This will kill the need for money of any kind.

Reply | Quote

Great Article and some great comments so far. The U.S. population cannot be cautioned enough about NOT using deep seek.

1 user thanked author for this post.

Slowpoke47

Reply | Quote

Only the US population?

cheers, Paul

1 user thanked author for this post.

rc primak

Reply | Quote

I just don’t accept that using readily accessible materials available on the internet (not password protected or encrypted) for training AI’s is theft, any more than a person using a library or browsing webpages to learn something should be considered a thief.

Reply | Quote

You need to review the extensive discussions (including here at AskWoody) about the legal issues surrounding Fair Use. There have been wars over content distribution by libraries through e-books, with publishers enforcing draconian limitations on “fair use”.

No, you do not have the legal right in the US or Europe, to wholesale copy entire libraries of books and publications, let alone whole art galleries and collections of images, videos and audio content, and retain all that content in  your own commercial database. You do not have the right to use all that content for commercial purposes without compensation to the copyright owners. This applies equally to content available on the World Wide Web, but owned by commercial interests.

-- rc primak

Reply | Quote

We are in the infancy of AI.  If you are impressed now, wait another 5 years!  Or even better, wait until we figure out how to do AGI.

Back in the late 1970’s I started my computer career working on an IBM 168, a machine with 2MB of memory and [supposedly] 2MIPS capability.  It took up a whole rather large basement of a building, required all kind of air conditioning and huge power cables, etc.  It cost $10’s of millions to purchase/(or was it lease only then?), feed and maintain.

Today’s cellphones that we hold in our hands have vastly more power and storage!  And even the most expensive models cost less than $2k!

The same reductions are going to happen in AI’s from size to power to $$ to operate.

And yet whatever happens, will we ever be able to develop the equivalent of a human brain, a 3lb biological package powered by the energy derived from the breakdown of organic food items and even junk foods?

Reply | Quote

AI as the term is now being used, is limited to LLMs. There are other AI applications, which do show the kind of promise you foresee. But LLMs are a dead-end street. They don’t have any actual intelligence, in the human sense. They just aren’t built that way. General AI is decades from posing any serious challenge to general human intelligence, if ever. Expert systems are limited to very specific use cases, where certain efficiencies of machine learning outperform human abilities. But these are edge cases so far.

-- rc primak

Reply | Quote