AI Safety Pipeline: A little more than accuracy scores

PyData Ireland is happy to partner with AI Safety Dublin in bringing this meetup to you.

Session Details:

The session aims at taking an overview of the challenges faced by Responsible AI Model Developers and precisely why 'Safe' AI is still very much a work in progress aspect of industrial SDC when all things are said and done. Some of these current challenges range from critical systems failures, bias, and AI-enabled surveillance, as well as emerging risks like technological unemployment, digital manipulation, and weaponization.

For some pragmatic exposure, we will also be discussing popular libraries for multimodal input (mainly text and visual) that incorporate red teaming and adversarial testing mechanisms for LLMs and VLMs. This exercise is to better understand the key considerations to remember while evaluating the pros and cons for your Safe(r than before) AI pipeline.

Finally, we spin up a few model testing instances on Colab using libraries like Text Attack and Inspect AI for a hands-on exploration and discuss the results (time permitting). This part of the session is aimed at creating a plug-play-and-experiment environment for the participants, and help them understand the finer nuances of developing a responsible AI pipeline (instead of only focusing on generating the mountains of AI governance paperwork).

Speaker Details:

Ojasvi Gupta, Founder of EquiAI Labs and AI Safety Dublin

Ojasvi has worked in the AI Strategy space for over 7 years with consulting enterprise firms like Accenture Strategy, bootstrapped startups and government think tanks like Mindfire Foundation(Lab42), where her work encompasses AI Governance, User Behavior and System Design Thinking.

She's been chosen for the founding cohort of Plaksha University's Tech Leaders Fellowship in 2019, and has also been a Forbes Ignite 2021 fellow where she worked on AI for Mental Health Diagnostics tool, she is a YLAC 2023 fellow working with a Member of the Indian Parliament (Dr Shashi Tharoor) in the DPI Ecosystem for international cooperation strategies. She's completed her master's in Human Centered Artificial Intelligence from Technological University Dublin and loves cats.