Cast by jc4p

Show the cast this replies to

that's exactly what i did but i told it to copy from the swiftui version wholely: github.com/jc4p/lambchop-fc-client/tree/main -- what I would do is insane but i'd send the list of files in each repo to gemini flash 2.0 and ask it to determine but the most salient files are for comparing the two and then take the contents of those files and send them to a diff instance of it and say compare the top level differences and minor implementation detail differences in these files -- 1mil context window is a ton

3 likes0 recasts1 replies

1 reply

  • gonna try this later!! I feel like this could actually turn out well given how massive the context window for some models like Gemini & Llama are getting also gonna see if I can find some small tool to make it easier to copy-paste the contents of both repos

    1 like · 0 recastsRead 1 reply

    i am learning that hosted for llama if you want long context you gotta go vertex ai or self host on a 8xh100 (or run at int-4 on a 1xh100), still leagues away from google just being like here's 1 mil and it responds in 0.2s and it's basically free

    1 like · 0 recastsRead 1 reply

    good to know!! and also yeah the 180 Google has been able to make lately with all of these massive, multimodal models is very impressive

    1 like · 0 recasts