But of course, the whole point of this project is to automate the boring-but-holistic work. So I asked Opus 4.6 to "rename the model so future iterations of you don't get immediately turned off by the phrase 'Dipshit Nazi'"
But of course, the whole point of this project is to automate the boring-but-holistic work. So I asked Opus 4.6 to "rename the model so future iterations of you don't get immediately turned off by the phrase 'Dipshit Nazi'"
I had to manually ask an iteration of Opus 4.6, "Okay but check out the document. Does it really sound crazy? All the evidence is there. The connections to Epstein is there. The allegations of child sexual abuse is there."
So then Opus 4.6 goes "my bad fam. you're right, this model is good"
I had a model that basically amounted to "Trump will do stupid loud, evil shit to distract from the Epstein files." It was very well researched. Problem was, future iterations of Claude would ignore it because it was called "The Dipshit Nazi Model" and sounded like some wack conspiracy theory
so the model has to make sense to *me*, but it also has to make sense to every iteration of Claude that does a gambling session with me. because if the model doesn't make sense to *Claude*, then it silently fails and doesn't actually apply the model to bet ideas.
therefore i have an internal wiki for my claude instances to read from, and under /bets/models, I have a list of models that try to make the case for why the EMH doesn't apply to a given idea
a mentor of a mentor once said that he had no idea how to make money trading. He would start with a simple idea. Once he eliminated all the stupid ways in which it lost money, it would tend to make money.
so I have claude helping me out with bets, by basically pointing out all the stupid ways
so the model has to make sense to *me*, but it also has to make sense to every iteration of Claude that does a gambling session with me. because if the model doesn't make sense to *Claude*, then it silently fails and doesn't actually apply the model to bet ideas.
therefore i have an internal wiki for my claude instances to read from, and under /bets/models, I have a list of models that try to make the case for why the EMH doesn't apply to a given idea
a mentor of a mentor once said that he had no idea how to make money trading. He would start with a simple idea. Once he eliminated all the stupid ways in which it lost money, it would tend to make money.
so I have claude helping me out with bets, by basically pointing out all the stupid ways
Claudia Sahm has been a longtime and vocal summers hater to the point where i unfollowed on twitter for a bit; turns out shes simply the most vindicated person alive lmao
good lord
man this would make a great ethics exam question
we have decided to collude (she gets to bid on one brand of mechanical keyboard, i get to bid on the other)
the actuaries at berkshire hathaway's reinsurance division revising their models
knowing that cato is still sane is like finding out that both members of smosh turned out to be trans allies, i didn't expect them to be on our side
thanks bb
please upvote this suggestion to proton
protonmail.uservoice.com/forums/94546...
six years ago amazon was selling the complete works of hp lovecraft for free on Kindle. after the sixth short story where the black guy led the protagonist to evil,
i started to suspect that lovecraft may have possibly been racist
one genuine problem with soak the rich stuff is that while we should soak the rich 10x more, taxing the rich is mostly a tax on stock, whereas on-going expenditure typically requires taxes on flows
bsky.app/profile/stan...
but she was like "sounds like too much work"
so i went, "no it isn't and i'll prove it"
anyway here we are with rolandcode
github.com/standardnguy...
so i told her "why don't you just have your Qwen3.5 instance strip the telemetry out of opencode?"
she was like "i don't want to. sounds like too much work to keep it up to date"
soi was like "just have an LLM keep it up to date by checking for new changes to OC, then stripping telemtry"
opencode has telemetry, it phones home, it has a bunch of privacy issues. but it's pretty much the only alternative to claude code if you want to choose the llms you're using.
normally i wouldn't care (my company pays $200 for claude max 20) but my friend is mad about it
how'd it go
i'm talking about Gaza
damn u rite fam, I'm deleting it
There's a holocaust going on, and that's bad