Investing

[Screwball]

26 minutes ago, Hongbit said:

In 1984, James Cameron told us exactly how this all plays out when he made The Terminator.  

I suppose I watched that at some point. Was never a Arnold fan, but I get the point. 1984 fits too. Giggle...

They have been working on robot people, dogs, whatever, for quite some time. They are a machine and they are trying to now teach them to think. Once they get good enough, the possibilities are endless. Then all you need is to fit'em up with killing devices. Probably great a crowd control device. Where does it all end.

Not 1984 or the terminator, but another warning from a smart dude back in 2000. Bill Gay, founder and chief scientist of Sun Microsystems wrote an article for Wired magazine called Why the Future Doesn't Need Us

Very long and interesting read. I was in IT around then and used a Unix Sun station. Even talks about crazy Teddy K the Uni b o m b e r guy. <-- Waves to the spooks. 🙂

Edited January 27 by Screwball

[Deleterious]

Relax guys.  I asked AI if it was going to kill all the humans and it said no.  

[ewsieg]

9 hours ago, Screwball said:

Very long and interesting read. I was in IT around then and used a Unix Sun station.

I miss mine, what a great machine.

[Deleterious]

Typed this into Deepseek.  You can see it visually start to type out a reply when it disapears and gives you this answer.

Quote

explain the tank man photo from Tiananmen Square

Sorry, that's beyond my current scope. Let’s talk about something else.

I tried the following and got an answer.  So it will explain photos.

Quote

explain the blue marble photo

 

[oblong]

ah the blue marble photo.  Nice touch.  I appreciate that.

Someone said ChatGPT got Temu'd.

 

[Screwball]

Tech is getting hammered. NVDA down over 12%, MAGS down 2.7%, and the NASDAQ almost 2.5. They are all working their way back up so this doesn't look too serious at this time. Bonds were bid but they are now coming back some as well. Good, I have some bills to buy tomorrow.

[Tigeraholic1]

1 hour ago, Deleterious said:

Typed this into Deepseek.  You can see it visually start to type out a reply when it disapears and gives you this answer.

I tried the following and got an answer.  So it will explain photos.

 

Whelp

 

[gehringer_2]

12 hours ago, Screwball said:

I'm old enough to remember the days before computers. I was in IT, engineering, CAD, etc. at a high level through the entire transformation.

This AI stuff scares the **** out of me.

if you want to get really freaked about AI, read a book entitled "Nexus." I think the guy who wrote is a little over the edge with a lot his worst case scenarios, but it is stuff people better start thinking about.

[Deleterious]

 

[gehringer_2]

1 minute ago, Deleterious said:

 

Stock twit doesn't understand the real implication. It's not the HW, it's the open source SW. If it was on GitHub, there are probably already hundreds of thousands of downloads.

[Deleterious]

I actually think they have it right.

People look at this as a bad thing because AI companies are going to be using less chips.  But they miss out that now training an LLM is possible for more people because of the lesser hardware requirements.

So instead of 10 companies using a million chips each for 10 million total, you have 100 companies using 500K chips each for 50 million total used.

[gehringer_2]

33 minutes ago, Deleterious said:

I actually think they have it right.

People look at this as a bad thing because AI companies are going to be using less chips.  But they miss out that now training an LLM is possible for more people because of the lesser hardware requirements.

So instead of 10 companies using a million chips each for 10 million total, you have 100 companies using 500K chips each for 50 million total used.

It may also mean that if you are already sitting on a hot video card you don't have to pay some one else at all. If a centralized compute model can give you an answer in 10msec with 100K chips,  then can I  get my answer in a minute with a hot video card, of which there are already millions out there? There is a constant push and pull in computer tech between centralized and distributed computing models, and it's always cost that decides where the division ends up. We see a lot simple applications types like word processing moving increasingly to cloud models, but other more compute intensive apps like cad and math (e.g. solidwork and MatLab) still run mostly on local dedicated processing power. While yet other tasks like protein folding run on distributed computing models. It seems each compute type has an optimum and possibly unique solution model.

So I agree, the implications are not that bad for NVidia, because the processing is still going to be done, but before this week the projections might have been over optimistic to the degree that some part of AI gets done with chips that are already being sold into their established console computing market. The bad news is for  MS/Amazon/Google etc who thought that only they would have the scale to exploit the potential market.

 

 

Edited Monday at 05:47 PM by gehringer_2

[Screwball]

I want to know who bought tech and MAGS puts last Friday. Someone always knows. Guessing the Deepseek guys were short.

[gehringer_2]

Watching this morning, the other thing that is a little frightening is how easy a state actor like China could be running a hoax. What if DeepSeek doesn't actually exist or whatever it is isn't what China is claiming its. with the markets as volatile as they are on the least news, if the PRC had been selling short on tech and then dropped DeepSeek, they've have made their trillion before anyone figured it out.

EDIT: HAHA - I see SB is already thinking along the same lines!

Edited Monday at 05:51 PM by gehringer_2

[Screwball]

1 hour ago, gehringer_2 said:

if you want to get really freaked about AI, read a book entitled "Nexus." I think the guy who wrote is a little over the edge with a lot his worst case scenarios, but it is stuff people better start thinking about.

What freaks me out is all this power in the hands of idiots, and we have no shortage of educated idiots.

[Screwball]

NVDA is down almost 17% as I type this and has went from $140+ last week to $118 right now. Ouch! There is a little resistance around 114, but if it goes through that looking around 101. The next hard resistance is around $97.4 where a gap filled back in August. Gap was from May.

[Tigeraholic1]

If you have not sold (I did two weeks ago) NVDA ship has left port. Fun while it lasted...

[Screwball]

Pretty good and free article from IBD on NVDA;

DeepSeek-Driven Crash Makes Nvidia Investors Face 8 'Secrets' Of Selling. No. 2 Is Key.

[Deleterious]

In meme coin news.

President Obama's half brother launched and rug pulled a coin called $Obama.  

This one might be my favorite.  The Cameo angle is brilliant.  

Dean Norris’ X account hacked by scammers selling meme coin named ‘Dean’

 

 

[Deleterious]

Apple is the big outlier today, up almost 4%.

They do AI stuff.  But for a company their size and being in tech, their spending is relatively low.  Looks like investors are rewarding that today.  

[Tigeraholic1]

.

Edited Monday at 11:42 PM by Tigeraholic1

[Screwball]

Interesting day, and Cramer is still in front of a TV camera. Wild how the Dow ends up almost 290 points while the S&P and the tech heavy NASDAQ got whacked.

As stated above, AAPL held up, NVDA not so much. Dow 30 heatmap;

Looking at chart of the same, what a huge move during the day. Gap down open, takes a huge **** to 44026, then decides to go up all day to close at 44713, hitting 44727 just a minute before. That's a 687 point swing. Cha-ching!!!!

And the NVDA chart. This is a one year chart. I would we watching the 97.4 if it keeps going the wrong way.

Edited Tuesday at 02:20 AM by Screwball

[gehringer_2]

so back to the question of why DeepSeeks release hit NVidia so hard if the system still uses NVidia HW - according to the report below, DeepSeek is achieving their results with about 10% of the HW for training that previous systems have used. DeepSeek is "only" 2,000 GPUs. So one interpretation is that they have demonstrated that prior estimates of the HW investment to meet whatever the assumed market was can be scaled back by close to an order of magnitude. And of course this wll also ripple out the energy forecasts and that planning as well.

https://www.nytimes.com/2025/01/27/technology/what-is-deepseek-china-ai.html

Quote

The world’s top companies typically train their chatbots with supercomputers that use as many as 16,000 chips or more. DeepSeek’s engineers said they needed only about 2,000 Nvidia chips.......

DeepSeek’s engineers said they needed only about $6 million in raw computing power to train their new system. That was roughly 10 times less than what Meta spent building its latest A.I. technology.

 

Edited Tuesday at 04:19 AM by gehringer_2

[Deleterious]

Higher prices should calm the markets.  

 

[Deleterious]

I wonder how much of Nvidia's sell off the other day was done by AI trading bots.  Talk about ironic.