seva k

ml @ somewhere, kaggle expert. world modelling, end-to-end ai enjoyer

[seva k @seva_kg]

@ktoya_me had so much fun on this trip - thank you boris and claude for organising!

[seva k @seva_kg]

@ktoya_me boris is in his gaussian splatting obsession era ✨

[seva k @seva_kg]

@ktoya_me @AlexBracken1 @aiDotEngineer jokes aside luton airport is actually great given 21 min train to kings cross

[seva k @seva_kg]

@digitman_ dropping max reasoning mode during the talk, spending no tokens on visual perception

[seva k @seva_kg]

@iclr_conf 🇧🇷 reflections: if i were to attend only one talk, it would with no doubt be one from @jaschasd as part of post agi workshop.

when the author of diffusion calmly lays down the uncertain future that awaits us, and how we are the ones to shape it, you better listen!

[seva k @seva_kg]

yet to see a flat tire here, unlike the usual struggle to find a fit-for-ride rental bike back home

[seva k @seva_kg]

pov: you visit nyc as an avid lime bike user from london

[seva k @seva_kg]

@ktoya_me @aiDotEngineer biggest lime promoter on london twitter!

[seva k @seva_kg]

2026 is both "agi deployment" departments at frontier labs and hiring for SLAM engineers

[Bruno Santos🇵🇹 @brunoeducsant]

OpenAI is hiring for SLAM engineer. Who would say.

[seva k @seva_kg]

@ktoya_me can't believe you did all this just because of dog 💩 in a lime basket!

[Boris Starkov @ktoya_me]

i pulled 3 years of my lime bike rides in London and fed to claude

after visualising, it proceeded to identify: where i live, where i work, when i moved, when i changed job, when i had a gap, where my gym is, where i go for saturday brunch

(all based on my ride history only!)

[seva k @seva_kg]

@mkurman88 token pollution is real

[seva k @seva_kg]

@ktoya_me photo credits?

[seva k @seva_kg]

@holynski_ this post is such a give away of what internal representations genie-3 is using 👀

[seva k @seva_kg]

@ktoya_me sneak peek into the future of human <-> technology interaction!

[seva k @seva_kg]

fun fact - compute needed for running these reward models is very efficient, distributed, and shaped by millions of years of evolution. it's called human brain.

[seva k @seva_kg]

as more often folks look at AGI as smth that has goals rather than imitates actions, we ignore "glimpse into AGI" thats already here. its your feed recommendation algorithm - online RL system that maximises time spent on the app / revenue, continually learning user's preferences.

[seva k @seva_kg]

abundance of reward signal, provided by half a billion agents DAILY

[seva k @seva_kg]

@ktoya_me sf in a nutshell

[seva k @seva_kg]

sam altman saying “across the board” lives rent free in my head