Hadas Kotek 🦄

@hadaskotek.bsky.social

Linguist in tech | Responsible AI, NLP, data science | AltAc advocate | mentor | cat lover | feminist | currently: Apple/MIT she/her | grayspec | 🏳️‍⚧️ ally | BLM | powered by coffee & spite | 📍 Cambridge MA שַׁלַּח לַחְמְךָ עַל פְּנֵי הַמָּיִם כִּי בְרֹב הַיָּמִים תִּמְצָאֶנּוּ

one thing about writing an entire book about sandwiches is my brain is just full of random facts now. coffee plants were forcibly introduced to java by the dutch. tuna fishes have extraordinarily high lung capacity. barbecue was a status symbol denoting slave ownership in the antebellum south

Many, many years ago I worked for a certain legal news website to which Lindsey Graham subscribed. Anyway i hardcoded it to always spell his name wrong in the top corner, no matter how many times he tried to manually correct it.

Today I talked to my Intro Linguistics students about bias in LLMs. On the day of a lecture like this I poke around to see the various ways that current models still should not be trusted. Even on basic things. Last time I gave a talk on this (a few months ago) here's what Llama said. 1/9

Screenshot from Llama 4 Scout chatbot with the prompt "How many r's are there in the word 'strawberry'?" and the response "There are 2 r's in the word 'strawberry.'"