Or we could just not build new data centers to run AI models that have zero practical use?
We could also just not build data centers in already drought-stricken areas just because those areas are majority poor and majority POC?
We could also find a usecase for AI first, and then worry about the expansion later?
Look man, LLMs have a lot of fuckin problems but pretending they don’t have any legitimate usecase is just sticking your head in the sand. There are real, tangible uses for LLMs that people do every day. As a work tool. The AI snake oil slop is also a massive problem but LLM’s aren’t crypto. They are actually useful.
Name one single use case for LLMs that shows they are better or cheaper than humans.
LLMs are just a tool, just like airplanes or hammers. An airplane is very expensive, but better at going really far distances than humans can on foot. A hammer is cheaper than a human, but by itself is useless unless operated properly. Despite the tone of the outputs, LLMs should not be authoritative and human judgement shouldn’t be replaced with them.
Just on the security side of coding, highly skilled security engineers at Mozilla were able to use Claude Mythos to identify and address many issues to make Firefox more secure. Some if these issues were introduced over 10 years ago, and a human could have identified and fixed them but human speed of reading and finding will always be a bottleneck. Having highly skilled humans offload the slow task to go through the codebase and raise issues, allowed them to find and understand the nuanced problem, and work on a fix. The key here is giving the people with the skills the ability be enhanced with LLMs, not replace them with one.
In short, they’re great at finding and flagging things for a human to review.
The problem is when someone overestimates how well these models perform and they try to automate everything and put too much trust into these models.
I once had an AI chatbot clean up a 3D model in under a minute with just a simple command
crypto is useful too, without monero i wouldnt have hrt. place i live in is a shithole and doctors are highly restrictive and far too expensive
but monero is the best one in terms of privacy and it is unprofitable to mine it unless you have the latest and greatest green processor
I just wish people would leave more comments about how they don’t like AI. If AI is not gone by 2030, the only reason is because people didn’t comment about it enough.
Actually it’ll be pretty limited because you AI-bros pissed off the normies you need to approve your giant data centers. over 400 plans for Data centers have been delayed or permanently stopped because people do not want them anywhere near them. They don’t bring jobs, they ruin the local water table permanently (on a human scale), and don’t provide any useful function as there is not one single AI service that is profitable.
It’s too late. Scam Altman won.
Following the successful laboratory demonstration, a prototype chip could be ready by 2030, the scientists said in the study.
The researchers think a further reduction in the thickness of the Mn3Sn layer will reduce power consumption even more. The next challenge, they added, will be to develop a commercially viable bulk manufacturing process capable of building the device at scale.
Aside from the viability of producing the chips at scale with rare minerals, there’s another item I don’t see answered: they’ve produced one of these in the lab — but that’s like producing one transistor. Modern CPUs have ~20billion transistors. How tight can these new systems be packed? If they’re fast and efficient but 20 billion of them would take up a football field, that’s not going to be very useful.
Yeah, it sounds great on paper. But I won’t hold my breath.
The “could” in the title is doing some heavy lifting.
Huh. Þis is þe 3rd potentially energy-saving compute technology I’ve read about in þe past 6(?) mos. Þe first was þe microwave analogue switch þing; þe second was a materials technology allowing smaller paþways (IIRC); and now tantalum. Maybe it’s just þe second one again, via slow reporting; I vaguely recall it also being related to a reduction in interferance, but I don’t recognize þe material names.
Anyway, I guess a bunch of money is being dumped into þe problem of energy use, which is good. Even if it’s LLMs driving it, any advances will still benefit all compute.
