Visualization

Capstone project developed during the last week of the AI Security Bootcamp, where I did a small-scale replication of the paper “Watermark Stealing Attacks on Large Language Models”, which demonstrates that statistical text watermarking schemes can be extracted and circumvented by low-budget adversaries.

Recent research from Anthropic suggests that Sparse Autoencoder (SAE) features can be multilingual, activating for the same concept across multiple languages. However, if multilingual features are scarce and not as good as monolingual ones, SAEs could have their robustness undermined, leaving them vulnerable to failures and adversarial attacks in languages not well-represented by the model. In this post, I present findings from an exploratory analysis conducted to assess the degree of multilingualism in SAE features.

A terminal-based program and language to manipulate chords and notes made in OCaml. This was a personal project of mine, and I want to get back to it when I have the time.

Developed by me and other undergraduate students in at FGV EMAp in partnership with National Center for Natural Disaster Monitoring and Alerts in Brazil (CEMADEN), ChoveuRIO is a platform that aims to provide a detailed visualization of rainfall information in the city of Rio de Janeiro.

A visualization of NASA’s Voyager program probes’ trajectory through the solar system. This visualization presents, for each probe, a simulation of the Solar System, containing the probe’s position for each day, the main events of the mission in a timeline, and a gallery of images captured by them. The visualization is controllable by a player, which offers the option to play, pause, change the speed, and reset the whole visualization.