OT: Stock trading

[B1GEast]

AI is like when the internet started, and put it on steroids.

I think the winners will be those who can use AI and not use a disgusting amount of energy.

They're all investing in nuclear power. The LLM,'s require constant power. Somehow Deepseek supposedly has found a way around this

 

[kobebeanbryant]

Regardless, no way I would let my companies data run on an LLM from China.

If it is an open source model that you can self host, then what risk does that pose to your companies data wrt China?

 

[temery]

As of this morning I am $1886 from my retirement goal (I retired ten years ago).

Prediction: The market crashes today.

 

[HuskyHawk]

If it is an open source model that you can self host, then what risk does that pose to your companies data wrt China?

None if your people have checked the code and continue to check the code for vulnerabilities. It's a trust question. But no different than relying on any open source.

But, per the thread I posted, they mostly achieved efficiency by (a) not trying to ingest everything, instead focusing on specific areas and (b) using a speed reading approach that looks at phrases not words.

Some areas may not be comfortable with the 90% is good enough approach. In other areas it will have no impact.

 

[HuskyHawk]

They're all investing in nuclear power. The LLM,'s require constant power. Somehow Deepseek supposedly has found a way around this

 

[B1GEast]

View attachment 106655

With over a billion people and being a large producer of manufactured goods, their power consumption should be far greater than the US.

We still don't know if they were achieved results by using US technology.

Energy Efficiency: The Real Reason DeepSeek AI is Such a Big Deal | OilPrice.com

DeepSeek, a Chinese AI startup, disrupts the tech industry with its energy-efficient and cost-effective AI model, sparking debate about the future of energy consumption and AI development.

oilprice.com

 

[HuskyHawk]

With over a billion people and being a large producer of manufactured goods, their power consumption should be far greater than the US.

We still don't know if they were achieved results by using US technology.

Energy Efficiency: The Real Reason DeepSeek AI is Such a Big Deal | OilPrice.com

DeepSeek, a Chinese AI startup, disrupts the tech industry with its energy-efficient and cost-effective AI model, sparking debate about the future of energy consumption and AI development.

oilprice.com

They did. Definitely diverted Nvidia from Singapore.

 

[Palatine]

They're all investing in nuclear power. The LLM,'s require constant power. Somehow Deepseek supposedly has found a way around this

China built a coal fired power plant every week in 2024 (50 of em). That would be a way around the DeepSeek electricity problem

 

[B1GEast]

They did. Definitely diverted Nvidia from Singapore.

I would not be surprised. I've used the Deepseek app. It rivals chat GPT 4.

With the new RISC 4 open source architecture for chip making, NVIDIA will have competition.

So even if they cheated, they've arrived and will have options

 

[B1GEast]

China built a coal fired power plant every week in 2024 (50 of em). That would be a way around the DeepSeek electricity problem

Deepseek supposedly doesn't a problem with electricity. From what they reported, and it's speculative, Deepseek only requires a quarter of the power that US companies need to power their language model.

Some people have downloaded Deepseek and got it to run on a raspberry pi device coupled with a decent video card.

 

[NJHusky]

I saw somewhere that they asked an earlier version of Deep Seek ‘who am I’ and it said it was chatgpt….

 

[Excalibur]

They did. Definitely diverted Nvidia from Singapore.

They've basically admitted it at this point

 

[HuskyHawk]

Deepseek supposedly doesn't a problem with electricity. From what they reported, and it's speculative, Deepseek only requires a quarter of the power that US companies need to power their language model.

Some people have downloaded Deepseek and got it to run on a raspberry pi device coupled with a decent video card.

It isn't the power to run it, it's the power to train it and keep training it. AI apps can obviously run on an iPhone or PC. It's training the LLM, ingesting essentially much of the data the world creates every day. That is what drives the need for AI factories and lots of power. Once you start, you can't stop or the LLM isn't up to speed. If you asked it when was the last U.S. airline crash and what were the details it has to give you a different answer today than it gave you on Monday.

Deepseek is not ingesting nearly as much stuff (much more targeted) and the approach to training it is a "good enough" approach. It's basically speed reading the data, where you look at sentences not words. Could possibly lead to inaccuracies but not many, so the efficiency trade-off is huge.

It's all pretty disruptive and exciting, but understandably a bit scary at the same time. It does remind me of the early days of the WWW. The internet was there, like AI was there, but until Netscape and the browser from Cornell Law School you couldn't do much with it. Usenet was fun for me, but so limited. We're at the stage with AI were Netscape is out, people are starting to build websites with actual pictures but there's no real ecommerce yet. I built a tourists guide to the Bay Area website back then, it was probably the first one. But I couldn't conceive of TripAdvisor. There are people with vision out there who will surprise us with AI.

 

[B1GEast]

It isn't the power to run it, it's the power to train it and keep training it. AI apps can obviously run on an iPhone or PC. It's training the LLM, ingesting essentially much of the data the world creates every day. That is what drives the need for AI factories and lots of power. Once you start, you can't stop or the LLM isn't up to speed. If you asked it when was the last U.S. airline crash and what were the details it has to give you a different answer today than it gave you on Monday.

Deepseek is not ingesting nearly as much stuff (much more targeted) and the approach to training it is a "good enough" approach. It's basically speed reading the data, where you look at sentences not words. Could possibly lead to inaccuracies but not many, so the efficiency trade-off is huge.

It's all pretty disruptive and exciting, but understandably a bit scary at the same time. It does remind me of the early days of the WWW. The internet was there, like AI was there, but until Netscape and the browser from Cornell Law School you couldn't do much with it. Usenet was fun for me, but so limited. We're at the stage with AI were Netscape is out, people are starting to build websites with actual pictures but there's no real ecommerce yet. I built a tourists guide to the Bay Area website back then, it was probably the first one. But I couldn't conceive of TripAdvisor. There are people with vision out there who will surprise us with AI.

I understand that the LLM's require a constant stream of power, which is why I indicated that US companies are procuring nuclear power plants.

A "good enough" strategy may be appropriate in some cases.

I agree that it's a lot like the early days of the Internet. People always say they remember what life was like before the Internet and will say the same about AI

The AI timeline is accelerating and soon things are going to get weird. How we work, with a transformer chatbox, how we interact with it and so much more.

Unfortunately, within a few short years, there will be less and less people employed in the corporate world.

I'm a bit pessimistic about how AI will disrupt our lives. It really is Pandora's box and when it's paired with robotics especially in the battlefield, who really knows where we end up.

 

[HuskyHawk]

I understand that the LLM's require a constant stream of power, which is why I indicated that US companies are procuring nuclear power plants.

A "good enough" strategy may be appropriate in some cases.

I agree that it's a lot like the early days of the Internet. People always say they remember what life was like before the Internet and will say the same about AI

The AI timeline is accelerating and soon things are going to get weird. How we work, with a transformer chatbox, how we interact with it and so much more.

Unfortunately, within a few short years, there will be less and less people employed in the corporate world.

I'm a bit pessimistic about how AI will disrupt our lives. It really is Pandora's box and when it's paired with robotics especially in the battlefield, who really knows where we end up.