Bought a sign company that was founded in 1961.
What I inherited: filing cabinets, handwritten estimates, customer history stored in someone's head.
60 years profitable. Zero data.
@yacineMTB The expensive part isn’t the hardware it’s the cost of training the ai.
Open sourced AI’s is literally giving away billions of dollars to people who have the hardware to run it.
Even 150 K is nothing.
@DavidRobinr0 The church accepts all types of people, including promiscuous women.
The church is also your best place to find a pure woman.
But do and believe what you want
The average new full-size pickup sold for ~$66K this spring. Full-size SUVs, ~$80K. Middle-class Americans finance that every day.
A $50–60K rig that runs frontier-class AI offline, that you own outright with no subscription, no ToS, and no kill switch, isn't a "business-level expense." It's a smarter use of the same money. People just usually point that $60K at a truck.
And that's the part that isn't cloud: cloud is renting forever, this is owning.
A year ago it was Musk-tier. The accessibility curve has moved a million-fold and only bends one way.
@Ready2ESC@jun_song I’m putting a 4 rtx 6000 pro server rack together for my business.
This will handle my agents, work flows all mid to low level ai needs.
I plan to have a return on investment.
If I was a hobbiest, would I need fable level ai? I don’t think so. Do you?
@Ready2ESC@jun_song ~$60K in GPUs vs. the ~$125B Anthropic raised to build it.
The hardware was never the expensive part. The development was, and open weights hand you that for the price of a truck.
Besides, hardware reqs will lower with each passing year. You’ll have this eventually.
Agreed.
The open source models are impressive.
However, it should be obvious that they will never surpass frontier models.
Closed models have full access to open models. Open models have limited access to closed models.
I am a supporter of open source but I won’t deny the frontier models.