Data Science & AI #14 - DataFrames for the Future & Training Deceptive LLMs

For this event, we will dive deep into Polars, a fast data frame library, that is a high-performance alternative to Pandas. We will also explore sleeper agents, i.e., deceptive language models, an interesting topic in AI safety research

This time, we will be back at Zollhof!

We will have exciting talks and our usual chitchat with drinks. This is your opportunity to meet other data science professionals from the Nuremberg area, sharpen your skills and exchange ideas with like-minded professionals.

Remote participation link: Microsoft Teams

Agenda:

1. 17:30 Welcome with Lukas Spranger and Benjamin Linnik
2. 18:00 Bernd Kaiser: Polars - DataFrames for the new era
3. 19:00 Óscar Nájera: Sleeper Agents – Training Deceptive LLMs that Persist Through Safety Training

---------------------------------

Bernd Kaiser: Polars – DataFrames for the new era

In this presentation, we will explore Polars, the emerging DataFrame library developed by Ritchie Vink. Our journey begins with a brief overview of Pandas - everyone's first DataFrame love - and discusses the need for an alternative framework. We will then explore how Polars, in conjunction with Apache Arrow, is redefining the standards for efficiency and performance in data analysis. In addition, we will examine custom extensions to Polars using Rust to extend its functionality while maintaining performance. Finally, we will reintroduce an old acquaintance that eases the transition to Polars, demonstrating its practicality and ease of adoption.

Bernd Kaiser works as a full-stack software developer at inovex. After studying computer science with a focus on IT security, he made a short excursion into the world of finance, only to return fully to the world of the Internet. In the process, he brought music effect devices into the cloud, roller coasters into the App Store and virtually united patients with their therapists. Many of his projects used JavaScript (later Typescript), and even after more than 20 years, there's still no end in sight.

Dr. Óscar Nájera: Sleeper Agents – Training Deceptive LLMs that Persist Through Safety Training
From political candidates to job-seekers, humans under selection pressure often try to gain opportunities by hiding their true motivations. They present themselves as more aligned with the expectations of their audience than they actually are. If an AI system learned such a deceptive strategy, could we detect it and remove it using current safety training techniques?

Dr. Óscar Nájera obtained his doctoral degree in solid state physics researching metal-insulator phase transitions by developing simulation and data analysis software to that end. His industry experience includes developing open source software for monitoring IT infrastructure. Furthermore he is also interested in blockchain technology, AI and economics.