Niko (@nibalic) 's Twitter Profile
Niko

@nibalic

🚀 run ai code @daytonaio, lord of mischief, university lecturer, pronoia cultist 🔮, forever falling down the rabbit holes 🐇

ID: 945223938518470656

linkhttps://www.daytona.io/dotfiles calendar_today25-12-2017 09:25:23

1,1K Tweet

506 Takipçi

838 Takip Edilen

Niko (@nibalic) 's Twitter Profile Photo

OMG, like when you’re all “bro, can you totally snap my pic real quick” and then you literally pass out mid-pose, as AI Engineer was so intense! Byeee New York, it was super fun and whatevs!

OMG, like when you’re all “bro, can you totally snap my pic real quick” and then you literally pass out mid-pose, as <a href="/aiDotEngineer/">AI Engineer</a> was so intense! Byeee New York, it was super fun and whatevs!
Niko (@nibalic) 's Twitter Profile Photo

Just checked some old 2022 research numbers, fun fact: the older you get, the more satisfied you are with your IDE. Wondering how things have shifted since then.

Just checked some old 2022 research numbers, fun fact: the older you get, the more satisfied you are with your IDE. Wondering how things have shifted since then.
Niko (@nibalic) 's Twitter Profile Photo

indeed the nano banana one is something else entirely Crystal ... feed it an older pdf of research I did. but it is just image gen from kimi created outline? though, it looks decent enough, maybe too flashy even, and not totally coherent

indeed the nano banana one is something else entirely <a href="/crystalsssup/">Crystal</a> ... feed it an older pdf of research I did. but it is just image gen from kimi created outline? though, it looks decent enough, maybe too flashy even, and not totally coherent
Niko (@nibalic) 's Twitter Profile Photo

Been playing with running evals. With proper function calling, gpt-oss-120b matches or beats OpenAI's gpt-5-mini in accuracy and kappa! However, gpt-5-mini still has better fail recall (86% vs 76%). Still if you are doing it often it makes sense also to go oss route cause of diff

Been playing with running evals. With proper function calling, gpt-oss-120b matches or beats OpenAI's gpt-5-mini in accuracy and kappa! However, gpt-5-mini still has better fail recall (86% vs 76%). Still if you are doing it often it makes sense also to go oss route cause of diff
Niko (@nibalic) 's Twitter Profile Photo

I just birthed a nuclear warhead of an MCP talk that’s gonna make the front row need therapy and the back row need a safe space in another zip code

I just birthed a nuclear warhead of an MCP talk that’s gonna make the front row need therapy and the back row need a safe space in another zip code