Oh my goodness, the needle is getting closer to the bubble at each passing day. Thank you for the good news!
O7
But this is always how computers tend to evolve. First you start with a giant centralized mainframe, and eventually you miniaturize it into a laptop or phone. Let’s hope these SLMs really do take off, for environmental and privacy concerns.
Exactly, we’re in the mainframe era of this tech, and it’s already coming to an end. I’m really optimistic we’ll see local models in wide use before long. Alibaba is already making chips specifically for that use case too https://wccftech.com/alibabas-tsmc-built-5nm-risc-v-chip-xuantie-c950-now-runs-qwen-3-8-27b-model-natively-unlocking-massive-vertical-integration-tailwinds/
Oh yes, what a bright future, open source ISAs, open source SLMs, open source agents (not there yet, but getting closer…)! Hope we get fully automated luxury communism at some point after as well.
and it will all be built in China too
I suspect we’re well into diminishing returns for the hyperscaling approach. The difference between frontier models now and three years ago is exponential. But between now and three to six months ago? Both are nearing their upper limits – throwing more money at them won’t change much, not without a fundamentally different approach.
That’s my impression as well. Seems like we’re reaching a plateau of what the current architecture can do. And it also becomes a race to the bottom. Labs have to spend more and more money to squeeze out ever smaller improvements, so it’s very hard to differentiate yourself from the competition.
I really just need openai or anthropic to either become profitable or fuck off. I am really sick of them.
My prediction is that one of them will become a vendor of record for the government and get a bailout, and the other will die off.
Oh my goodness, the needle is getting closer to the bubble at each passing day. Thank you for the good news!
