@OfficialLoganK@GoogleDeepMind Frontier of intelligence, speed, and cost - pick 2? would love if you can have a "flash" that optimize speed/cost. Gemini 2.5 Flash-Lite was a great model for it's cost ($0.1/m) - could you distill a newer model to have a "smarter" version, the same cost? like GPT-5 nano.
@cremieuxrecueil There's an argument to be made that maybe they are not aware enough and so it's "fine", just like maybe we humans are already factory-farmed at some capacity (like, for "aura" or "prayers" ala American Gods book/show) and we're not aware of it, so it's "fine" :)
@nateberkopec@charlieholtz @SuperconductDev SuperconductDev is remote right? I was looking into some solution that will work locally (with https://t.co/NiGMszSSa9 as the control interface) but maybe it's too complicated :(
Still really want to use a local git repository with worktrees that already there.. or at least without connecting to github... even after adding a local repo, it still tries to fetch origin (no! bad conductor! use local!) and I get this error:
"Failed to create workspace: Failed to fetch from origin: [email protected]: Permission denied (publickey). fatal: Could not read from remote repository. Please make sure you have the correct access rights and the repository exists."
😭
Thank you! it didnt work 😢 it still tries to connect to github even though I've opened a local git repo. Error msg:
"Failed to create workspace: Failed to fetch from origin: [email protected]: Permission denied (publickey). fatal: Could not read from remote repository. Please make sure you have the correct access rights and the repository exists."
@aviflombaum Sounds good - def getting something out is more important - and I might be completely wrong anyway :P
Love that you're doing it - looks amazing!
@aviflombaum Anyway we have our in-house db prompt objects have a model field too but we want to move away from that as we found it's not as flexible. I figure I should try to get you to think about that assumption given that you're super early. Happy to chat too if you're up to it :)
You're right - we do write prompts with models in mind. But isn't it the same for ruby code too? you write a ruby class and run it on ruby 3.0 or 3.3 - sometimes you write code that will only work on 3.3, or will work differently on 3.3 (because of a breaking change)
But you dont write on every class "only ruby 3.3" at the top. (only in a gemspec, and even that is more flexible i.e ~> 3.x)
@aviflombaum kinda how I can take the same html and render it on different browsers - the essence of the html is the same, even if it renders differently be different browser quirks.
@aviflombaum Why do you have the model as part of the prompt?
I can take the same prompt and run it on different models - the result might be slightly different, but the essence of the prompt is the same