Post

This is the take I agree with. The large scale LLMs are going to phase out in favor of local LLMs. Look at GPT OSS 20B and tell me it’s not good enough for 90% of use cases.

Autumn Nash 🎃👻💀🖤@withenoughcoffee.bsky.social

This is actually a cool and interesting take. timesofindia.indiatimes.com/technology/t...

timesofindia.indiatimes.comPerplexity CEO Aravind Srinivas says multi-billion dollar data centres will become less relevant and ‘biggest threat’ is… - The Times of IndiaTech News News: Perplexity CEO Aravind Srinivas has issued a warning: the future of artificial intelligence (AI) may not be in the cloud, which means that massive, en.

3/1/26, 1:15 am2 reposts6 likes

twotwotwo.bsky.social

@twotwotwo.bsky.social

3 Jan

I find it hard to make specific predictions but it's notable that smallish models (GLM, Haiku, maybe Minimax/MiMo?) can Actually Do Stuff according to both benchmarks and anecdotes

twotwotwo.bsky.social

@twotwotwo.bsky.social

3 Jan

And there're improvements in sparse attention, quantization, MoE efficiency, hardware. You could imagine it continuing until hardware meets it and local models really do handle a lot. But even if not, improving efficiency suggests you might not need 1000 1GW datacenters!

Frank Foster

@frankefoster.com

3 Jan

Yup. Even looking at tech over the years we went from room filling computers to your own machine doing so much. LLMs will eventually be outpaced locally no matter what.

tayarndt

@handle.invalid

6 Jan

We need to do is have these small models being trained for specific use cases like swift and swift imagine how good that could be.

Michael Doise

@michaeldoise.com

6 Jan

I talk about this a lot, but that is why I think Apple has the right approach with Foundation Models. There is a lot of potential for built in on device models to perform specific tasks. These models will be more useful in the longrun compared to expensive data centers.