about

i studied physics, and it left me one habit i cannot turn off: asking what is actually happening underneath. most of ai today runs on models nobody can fully open up and read. that is either scary or interesting, i picked interesting.

after physics i tried a computer science master's in padova. it did not work out, and i wrote about why on the blog, in turkish. the short version: i could not feel myself getting better, so i stopped. it was the right call, i still think about it though.

now my thesis at sabanci is on the refusal direction in llms. when a model says "i can't help with that", something specific happens inside, and it can be shown as a direction in activation space. i am mapping how jailbreaks move the model off that direction, and what that means for defense (or attack).

at SCL i build an ai based market research engine: synthetic consumers that behave like real ones, which is a strange sentence to write. days are product, nights are model internals. the two feed each other more than i expected, we will see where it goes.