Chat GPT or DeepSeek

[David]

I have no experience with Chat GPT. The enormous computing power required for AI has sort of worried me, but I have not been following AI closely.

Now the Chinese have their own more efficient AI. So far the biggest effect is on the Magnificent 7. This will affect me as I own ETF in my IRA.

Does anyone here use AI? For what?

[David - WI]

It seems DeepSeek is pulling a bit of a hoax. They used the expensive competitor's AI to train their model, so they can claim they only spent $6M "training" it... but apparently everybody else could do the same thing if they wanted to.

Also. you have only the word of the Chinese government that DeepSeek is more efficient and used less power... so far I haven't seen any "impartial" data.

Maybe it's legit... but that seems about as likely as the Chinese stealth fighter beating the F22 or F35.

China’s extraordinary claims about DeepSeek’s cost-efficiency have raised skepticism, particularly given its reported $5.6 million price tag for training the model—far below the estimated $100 million spent by companies like OpenAI on similar AI systems. While this figure has been widely promoted, industry analysts point out that it likely refers only to the final training run, omitting significant research and development costs. This selective reporting could mislead investors and policymakers, creating the illusion of a major breakthrough when in reality, the full expenses may be far higher.

Additionally, China possesses vast numbers of Nvidia GPUs, meaning the model may have been developed using the very hardware it claims to have bypassed. There is also speculation that DeepSeek leveraged outputs from existing U.S. AI models to optimize its own, potentially lowering development costs while raising concerns about the originality and independence of its work.

The lack of transparency surrounding DeepSeek’s full development cycle, its reliance on already established AI frameworks, and the broader context of Chinese propaganda efforts suggest that its cost-saving claims may be exaggerated. While the technology itself may be legitimate, its narrative appears carefully curated to advance China's geopolitical goals rather than presenting a truly independent and cost-effective AI revolution.

https://drrobertcastellano.substack.com/p/deepseek-chinas-propaganda-machine

[John S]

Another forum I frequent has some engineers that work for some automotive performance manufacturers, and they do some pretty neat stuff with it. They like the fact they can build models and stress test parts more efficiently, and it can adapt to create some pretty wild stuff.

I'm in construction,
I have to do a lot of procedural work before executing scope, and I have tried it on a couple of my MOPs (method of procedure), and I guess it was okay. It definitely saved time, but it also missed some key points. It's possible I didn't give it the correct inputs, but these things need to be air tight, due to the possibility of energizing, or not deenergizing in the correct sequence, etc. It locked on to the LOTO portion, and did not understand the sequence as I had wanted.

I'm not convinced, but not totally skeptical either. I think it's like many technologies, that will stumble it's way to a significant and efficient tool. It's a good thing to keep up with this stuff, at least if you want to stay relevant in the workforce. And this is coming from the guy that till more recently than I care to admit, would change Excel files to PDF, lol.

Podcast I frequent, put it this way. The US AI will be top of the line, but less inclusive. The Chinese AI will scale to be more affordable, and available to more people, but not as cutting edge. Seems aboot right if using the past as a predictor.

[Shaun K]

It seems DeepSeek is pulling a bit of a hoax. They used the expensive competitor's AI to train their model, so they can claim they only spent $6M "training" it... but apparently everybody else could do the same thing if they wanted to.

Also. you have only the word of the Chinese government that DeepSeek is more efficient and used less power... so far I haven't seen any "impartial" data.

Maybe it's legit... but that seems about as likely as the Chinese stealth fighter beating the F22 or F35.


https://drrobertcastellano.substack.com/p/deepseek-chinas-propaganda-machine

*supposidly* China used 50,000 Nvidia A100's that were accumulated before us restricted nvidia GPU' export. I seen a tour a few weeks ago of Elons new Texas AI node being built using NVidia's new H100 (50,000gpu's) and 20,000 of his own designed GPU.

A First Look at Elon Musk’s New Cortex AI Supercluster | Technology Magazine

NVIDIA GPUs: H100 vs. A100 | a detailed comparison | Gcore

NVIDIA H100 Enterprise PCIe-4 80GB — Vipera - Tomorrow’s Technology Today