playgroup - LLM security - how safe are these tools?

​Newspeak House are kindly hosting this iteration of playgroup. There's no cost. Spaces are limited so if you take a space and don't show up, someone else loses out - so please act in good faith and decline if you realise you can't attend.

​playgroups - this event series brings together a group of likeminded LLM users in a room to work for a day on a set challenge. We push the boundaries of what's possible, share how we use LLM tooling and help each other figure out 'what actually works, right now'.

​These are organised by https://www.linkedin.com/in/ianozsvald, where after 26 years in AI he's still trying to figure out 'what actually works' so he can help his clients make better decisions and generate valuable outcomes. Ian co-founded PyDataLondon (16k members over 11 years of volunteering), co-authored three editions of O'Reilly's High Performance Python & has helped data science companies realise millions in outcomes from high value projects.

​About the day - we'll spend the day working on LLM based security with hands on activities. Can it be secured? How are we vulnerable? Knowing that our work practices are safe is of increasing interest to many.

​We'll solve some LLM hacking challenges via the excellent Lakera Gandalf platform - a safe security platform that's designed in levels for us to crack our way through, so we learn about well known weaknesses in chatbots and agentic systems.

​We'll prod at LLMs after abliteration (with their guardrails removed) to see what they know that they usually can't discuss.

​We'll discuss safe code development practices (like sandboxes). You'll end the day with a stronger idea of weaknesses and safe approaches for using LLMs in products and during work.

​Bring your questions, a laptop and a curiosity about this brave new world.

​This will be the 10th playgroup (https://playgroup.org.uk/) organised by https://www.linkedin.com/in/ianozsvald

​About you - you're an active user of LLMs, you have a curiosity about whether they can truly be made safe and you'd like to explore these questions in the company of a likeminded bright group of colleagues.

​Code of Conduct - we follow the Bill and Ted's "Be Excellent to One Another" doctrine. This is a professional space, be friendly and helpful and professional.

​You'll also be invited to the playgroup slack, a private slack full of smart and interesting LLM developers.