True Positive Weekly #174

Hey, in this issue: How the LLM watermarking works; Compression is prediction; Should researchers write papers for AI instead of people?; Putting sign language AI into users’ hands; What happens inside an LLM when it tries to calculate with nothing but matrices; An offline translator small enough for a Raspberry Pi; and more.

This newsletter has no ads. Please support my regular work on curating this newsletter (and my work as a book author) by becoming a Cool Supporter 🚀

• [Nature] How the LLM watermarking works

• [Explained] Compression is prediction

• [IEEE] Should researchers write papers for AI instead of people?

• [Google] Putting sign language AI into users’ hands

• TutorMoments: Do AI tutors know when to help and when to hold back?

• [Project] A calculator, compiled into a transformer

• [Explained] Arithmetic without numbers: What happens inside an LLM when it tries to calculate with nothing but matrices.

• [Project] gemma-translator: An offline translator small enough for a Raspberry Pi

Like or comment to help the newsletter grow.