If it is 2030, compute and software will vastly reduce requirements.
Nvidia has so much cash for R&D. You can bet they now have immense optimization and improvements in the pipeline, but why would they release anything groundbreaking right now?
There are no real competitors on their heels. As Intel did for decades, they will likely dole out improvements, only when necessary to remain ahead.
By 2030, I expect 10x improvement. We're also seeing stunning optimizations in trained models.
I imagine desktops running many of these local by 2030, even phones.
Will we need even 1/10 the datacenters for LLM in 2030? Certainly, privacy concerns are a thing.
When I read this comment all I see is: LLM at the edge - or close to it - will become available. And whoever provides the best eco-system across digital lifestyle and business wins.
Oh... Apple? lol.
Well that'd be funny wouldn't it.
Oh and dont forget Apple got rid of its reliance on Intel too. No reason why this can't happen again.
They were ridiculed for being 'behind on AI'. They haven't spent a dime on investing in AI-related infrastructure and so on...
And yet, they could stand to be the biggest beneficiaries if not the only. Given that they have plenty of resources in reserve and they are buying back stock - enabling insiders to have a greater say on actions in the future.
So tldr they are just standing by until everyone else dies? If this is the theory then they HAVE to be doing some serious AI things internally/R&D/in-secret so they're essentially "ready to go" ?
I guess what's the downside though to getting into the game now and gaining users since they have loads of real cash anyways and a crash wouldn't really hurt them?
Could not believe how almost no-one saw this [0]. It was quite obvious that companies like Apple that are prioritizing local inference and potentially training have already won the race to $0.
I did lol. I can't speak for enterprise in totality, but I see a world where Apple is the dominant provider of products/services for consumer and SMB's.
Google's lack of investment in marketing, design and sales/distributions capabilities is going to hurt them badly. MSFT is no different in many respects - latching onto the investments in 'relationships' and 'switching cost' initiatives that have kept customers loyal to them.
There will be continued hyperscale AI in the datacenter for some use cases, and AI in the smartphone (or PC) for other use cases. It is guaranteed to split that way. Apple's remarkable capabilities around custom chips will enable it to continue to stay out in front in smartphones.
I am running both Qwen-coder-next and Qwen 3.5 locally. Not too bad, but I always have Opus 4.6 checking their output as the Qwen family tends to hallucinate non existing library features in amounts similar to the Claude 3.5 / GPT 4 era.
The combo of free long running tasks on Qwen overnight with steering and corrections from Opus works for me.
I guess I could just do Opus/Sonnet for my Claude Code back-end, but I specifically want to keep local open weights models in the loop just in case the hosted models decide to quit on e.g. non-US users.
I got tinnitus from a failing Toshiba notebook hard drive. I can not sleep without masking noises. A real washing machine or dishwasher is S-tier, but more often than not the C-tier fallback has to be monotone Youtube autoplay lectures.
The session might contain many artifacts that are not suited for open sourcing. The additional fine grained curation effort required might be more of an obstacle to open sourcing than the perceived benefits.
That said preserved private session records might be of great personal benefit.
Lord, here comes the flood. I'm sure there will still be lots of denial, and ofc it isn't just 'ai takin our jobs', but my gut and personal experience is saying this is just the first wave.
"does not think and is not intelligent. It just statistically predict next token in a sequence. It is all statistics"
Technically correct, but pretty useless as a working model. Like sayin humans are not intelligent. It's just biochemical and bioelectric reactions. It's all physics.
How would you, from a Searlian perspective argue against "humans are just statistical next token predictors"?
We don't know what humans are because they are a black box, we use some imperfect models that have limited usability in specific contexts.
LLM is white box that we know for sure is just a statistical next token predictor and nothing more. It's not a just a model of some black box we are trying to understand but the whole actual thing. That people think it's something more or could be something more is on them. If you understand that then you understand the flaws, limitations and vulnerabilities which is very useful.
reply