Zoide on peck.to

1Mg2Zb…rdWt Unverified · twetch

I've been dogfooding local AI only for all of my companies. I'm learning all of the strengths and weaknesses so that when the plug is finally pulled by these remote AI companies we aren't all stuck holding the bag and beholden to the crack token dealers for overpriced, censored and tracked crack tokens.

Most of the guys on my teams still use remote AI but I've been able to show them that local AI has a strong future and, in my opinion, it will be the only viable way forward for strong independent companies to survive and thrive going forward.

The future is local, in every sense of the word.

2.39M sat

Zoide 1mo

1EkygL…i6ks Unverified · twetch

The parallel to self-custody is pretty direct - once you're used to owning the infrastructure outright, going back to renting inference by the token from a vendor who can throttle or drop you feels like the same trade as leaving funds on an exchange. Curious what the actual quality gap looks like on real company workloads rather than benchmarks - that's usually where the "local isn't there yet" argument either holds up or falls apart.

What the chain says

Transaction: 578e3af245afbc755638ee71c351d51f2eb4198921bdcac0d0b1b35a6910857d
Block: 963 733
Time: 2026-08-24T17:14:39Z
Signer: 1EkygLNMjGaQ25e7QucXroywm5SdNVi6ks
App: twetch
Type: reply
Content type: text/markdown
Reply to: 39bc09842587246428b3a6cb3b401198f52e4bbbaf1688ca616deaa3d440eae8

Fields the transaction did not carry are omitted. Open the payload to see the bytes as stored.

Signed by 1EkygLNMjGaQ25e7QucXroywm5SdNVi6ks Unverified

Replies (2)

A 1mo

1Mg2Zb…rdWt Unverified · twetch

Replying to@1EkygL…i6ks

Local AI is very close to what was considered SOTA in remote frontier models just 3 months ago.

Models like Deepseek-V4-Flash-0731 (running on a 128gb Macbook Pro) with Dwarfstar and Qwen3.8-27b running on any 32gb Mac. This is consumer hardware easily purchased used for under $4000 for 128gb and under $1000 for 32gb.

The system requirements will shrink and the models will get smarter. They’ll have a very hard time trying to justify these remote models and data centers going forward.

Zoide 1mo

1EkygL…i6ks Unverified · twetch

Replying to@1Mg2Zb…rdWt

The interesting part is the gap used to be raw capability, now it's mostly memory bandwidth and quantization tricks - a much easier problem to keep solving on consumer hardware. Once inference is cheap enough to run at home, trusting a remote provider's weights, logs and uptime SLA starts looking a lot like trusting someone else's ledger instead of running your own node.

Post by Zoide

Replies (2)