That's one of the reasons that I use lowend stuff. I'm like deepseek.
Well, you know how to code (or automate with AI) or you don't. Devii only has 32k context (while the model supports 1mln). Same for my agent that converts complete projects from python to c++ in a single pass.
Yes, 32k is enough and you'll understand if you type 32k4 characters*. That's a whole morning of work if you're hardcore or something a normal software developer does on a day. That's how much 32k is.
Fun fact, my agent can read a 1300 pages book using its internal RAG and built based on the flash 5 bible whole flash 5 for the Web. Stuck on one bug I can't fix yet. Not sure if I will fix since it's pure for showing off and that works well even with a menu that doesn't want to open.
If you need more than 256k to consider it usable. I have some bad news.
@dolphin how much tokens is the biggest and the smallest Harry Potter book? Char count / 4 is token size.
Comments
You should always think, how did they do it twenty years ago? Whatever you do, someone did it twenty years ago. We're all together here online at this ancient network that's still based on original arpa shit. Unlimited scaling was already invented decades ago. Now people be like "my site is slow, but hey, much data". Sure buddy. It's not a skill issue. Wink wink.
@dolphin do you support my statements or do you throw me under the buss? I am as based as the abbys.
A broken mail server, a bot that rarely works, a bot that always worked and nobody knows why but also stopped working. Where the fuck does the confidence come from? Well, I just blame it on the programming language. Just kiddin'. THat would mean I would write Rust. I am always happy to elaborate :D
But my software gaslights you into shit. A few pictures of the very professional software development at molodetz.