I tried an experiment. I had Tim do a refactor of one of my projects. It dockerized the whole thing (postgres, redis, node server, app) I used glm-5.3-flash. It took about an hour. The whole thing 9.25 MILLION tokens and cost me $0.39!
So, I crunched some numbers and found that this same operation would have cost around $36 using claude-fable-5.1!
Why are you still using frontier models that cost orders of magnitude more for the same result?
I want to apologize, in public, to @loftwah for comments I made last Tuesday. It was unnecessary, uncalled for, and uncouth. It was not my best moment. And you deserve better. I don't always agree with you, but that doesn't mean you are wrong. I hope you'll accept this apology.
For sale: weightogether.com
Domain name. Source code. 1 year of hosting. All the documentation.
It just sits there right now. If someone wanted to maintain and promote it, it might be worth something.
DM or email andrew@erwininteractive.com
Well, through some fault of my own, I find myself homeless again. The Midwest experiment did work out as I'd hoped and now I'm looking for a new place to land.
So, you might not hear from me as much as usual, but I'll do my best to keep up with email and DMs.
I need to raise $2,000 in the next 24 hours. So, if you need cloud hosting, code reviews, small projects completed.. let me know. I don't sleep, all work is done quickly.
ouch. Bloomberg reports that Microsoft only has 2 GW of AI compute running. Where are all those chips they bought from Nvidia?
Guys, more and more NOTHING about this industry makes any sense. And I think someone might actually go to jail before it is all over.