I've made some fun demos using the new kimi-k2-0905

Posted by Dr_Karminski@reddit | LocalLLaMA | View on Reddit | 22 comments

They were all created with a single-pass, AI-generated prompt using both claude-code and kimi-k2-0905.

[-]

Dr_Karminski@reddit (OP)

prompt:

https://gist.github.com/karminski/52a72d4726128c10a266bfb8270fe632
https://gist.github.com/karminski/0435b69c6d8c93b4bd1724b64e43bd75

[-]

power97992@reddit

Dude that is a pretty detailed prompt… Can you get the same results with a new simple prompt:” make a realistic simulation of many firecrackers exploding ” ?

[-]

eleqtriq@reddit

You’d be more impressed if it made this complex demo using only 8 words? Bruh. Being able to follow this complex prompt without missing anything is what makes this even more impressive. Not less.

[-]

eleqtriq@reddit

Imagine asking “I want a website” and LLMs spending all their over engineering a basic ask

[-]

When I do a detailed prompt I find it easiest to start with the basic core prompt and then have the LLM propose a detailed spec/PRD. From there you can twiddle the details and feed it back in a fresh session as something to build.

Aside from being a faster way to generate the wall of text, it also makes any iteration on the initial idea easier than the "start with a small prompt and go back and forth on changing the details after" approach - just modify the initial prompt for anything you might have missed/changed your mind on and you get an answer generated with a completely fresh and optimal context.

[-]

arman-d0e@reddit

Yea I feel like with a prompt like this, other models would also stand a chance of similar results. Nonetheless, K2 is a beast (even if it is fatter than your mom)

[-]