Artificial Intelligence thread

[Hyper]

on AI bubble, this is all you need to know. The economics right now is entirely ridiculous.

https://twitter.com/i/web/status/1956141182754677084

Hence my belief that Google will come out on top. They have full chain of chips, model and networking and datacenters planned out well in advance. TPUs are really cheap and gemini is crazy value for money.

 

[Wangxi]

https://twitter.com/i/web/status/1957107498055598568

https://twitter.com/i/web/status/1957179465584558360

 

[Fertorstar]

That requires building gas turbines to generate the power.

GE Vernova of Cambridge, Mass. is the world’s largest manufacturer of gas turbines and rapidly growing to boot

 

[tokenanalyst]

DeepSeek 3.1 Base is here:

 

[Hyper]

That requires building gas turbines to generate the power.

They are running aeroderivative gas turbine onsite.

 

[GulfLander]

I mean - the U.S. already has substantial compute and investors are going ga ga over electric utilities/transmission because AI has increased the returns on power infrastructure - not to mention that there are a substantial number of investors like insurers and retirees that do not aim to maximize risk and want predictable, steady returns that then are effectively an infinite demand well for standard infrastructure projects

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

see also
Transmission line construction boom

Please, Log in or Register to view URLs content!

Natural gas pipeline construction boom (alternative to long distance transmission since it brings fuel straight to the consumers)

Please, Log in or Register to view URLs content!

How long were the new transmission lines constructed recently? Are they also UHV?

 

[GulfLander]

Sam Altman admits OpenAI ‘totally screwed up’ its GPT-5 launch and says the company will spend trillions of dollars on data centers

Please, Log in or Register to view URLs content!

 

[gadgetcool5]

DeepSeek 3.1 Base is here:

View attachment 158723

This seems to be pretty much the same performance as R1 0528, except they cut the number of tokens needed by 50% and vastly increased the speed. While I think that is nice, it seems most people were expecting a bigger update from them and there is disappointment going around.

 

[Wangxi]

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

 

[tamsen_ikard]

This seems to be pretty much the same performance as R1 0528, except they cut the number of tokens needed by 50% and vastly increased the speed. While I think that is nice, it seems most people were expecting a bigger update from them and there is disappointment going around.

I think AI is moving towards more efficient networks achieving same performance instead of better performance. Cause it really hard to achieve better performance on chat bots without significant expansion in data quantity and quality.