And it’s open source…?
6 Likes
I’ll try to get it to work sometime this week
Woke up to this on the news. Free and open source is a big thing, for sure. But apparently no Nvidia chips were used in the creation of the Deepseek LLM. Financial experts are keeping a close eye on the Dow Jones Industrial Average and the Nasdaq Composite as many seem to think this might cause an “A.I. bubble burst”. I guess time will tell.
1 Like
11 Likes
Oh, they realized that “brute-force” approaches which relies on sheer computation power could be achieved in a similar scope with less computation power ?
Try to explain the halting problem to a day trader. And that it is de facto stupid to bet on a solution to the P = NP vs. P ≠ NP problem.
The value of my “diversified” retirement fund has suddenly lost 20% of its value. Apparently my fund administrator didn’t do his job and I’m over-exposed to AI-meme stocks like Nvidia, Micro$hit, Apple, and Meta. How did that happen?
My current portfolio looks a lot like this stock (names not real, but there dozens of products like this):
Fund 1: S&P 100 ETF (Most valuable companies include Nvidia, Apple, Meta, Alphabet)
Fund 2: Information Technology ETF (Most valuable companies include Nvidia, Apple, Meta, Alphabet)
Fund 3: Global ESG ETF (Most valuable companies include Nvidia, Apple, Meta, Alphabet)
Fund 4: MSCI World ETF (Most valuable companies include Nvidia, Apple, Meta, Alphabet)
Diversification? LMFAO. Of course that’s just the equities - I also have a small amount of bonds. But not enough to offset the collapse in value of the top US AI meme-stocks.
2 Likes
Tangential to this topic, here’s a link to a freecodecamp tutorial I found interesting. Even if you’re not into python, the broader principles as to how AI & ML is approached might be of interest to some…
As of 11:00am CST, the DOW Jones is pretty much flat, but the NASDAQ is down more than 600 pts (3%). The next week or so will tell us more.
I’m glad that this happened, it shows that we don’t have to feed the monster that’s nvidia. They have started going down on an expensive road, both financially and environmentally.
2 Likes
already found the git an hour ago, will read through the readme after work
ollama serve than in another terminal ollama run deepseek-r1:14b or any model than can run on your hardware. The 14b works well on my 16 gb ram laptop running on the cpu. Going to try a bigger model on my desktop pc. It’s just great, it runs well just with a CPU.
2 Likes
Just gonna say this and then exit.
There is no such thing as “open source AI” coming from big tech because none of them have sourced the text and other media used to create their AI ethically or legally at first.
Notice the separation of “ethically” and “legally” because these are two different things.
All these companies and VCs used unethical means at first, then made changes to their legal documents to make it currently legal until/if the law catches up.
Nuff said.
4 Likes
Well, in that regard it tend to say that DeepSeek AI is most likely being trained on datasets of problematic origin.
1 Like
That is unfortunate, but everybody with a pension would have taken some damage. Financial advisors are in a bit of a bind, if they play it safe then people complain their funds are under performing. Those big tech firms were printing money for the last decade, but yes, it is a bubble…
2 Likes
Good article from the BBC:
I have had a brief play with it, seems pretty standard, OK for historical, factual searches, but cannot answer anything “current” and:
Don't mention
Tiananmen Square
4 Likes
I’m not able to run the latest version (deepseek v3) on my laptop. However, I tried out the v2 model and noted that it performed relatively well when commenting on historical and current (well up to the summer of 2023) affairs. However, it has some notable geographic blind spots all right.
I am at work, so was running the browser based version, I still struggle to see the point, it looks like their model only goes up to October 2023, and I can do better with a search engine 
The problem with cutoff dates is that there’s always some current events/history that is missing. This is a limitation of scraping the web for ”news” - and even then, history often gets rewritten.
I would think that a better evaluation of an AI model’s suitability (and reliability) would be had by accessing responses that are more focused on immutable information. 
1 Like