Title
Ratiocinor, ergo mente aegrotare possum
Speaker(s)
Diogo Sousa
Date and time
2025-10-24T17:00:00Z
Session type
Lightning talk (05 minutes)
Abstract
Large language models are all the rage now. Every couple of months they release a new version: bigger, bolder, wiser, capable of besting humans in any test and bringing us ever closer to Judgment Day.
Massive amounts of data are ingested to train these models, with few criteria other than “we can get our hands on it”. This, however, comes with its own set of hazards.
GPT-5’s system prompt was leaked almost immediately from payloads hidden among its training data by researchers like Pliny the Elder Liberator. A recent story on Google’s Gemini also brought to light some of those issues when it found itself caught in a bad loop: Google Gemini struggles to write code, calls itself “a disgrace to my species”.
A model is only as good as the quality of the data thrown into it … and there are fewer and fewer high-quality data sources to go around, courtesy of, well, using LLMs to generate most of it. What happens if you deliberately poison whatever you can find?
In this Lightning Talk, I will mix philosophy with wanton aggression in pursuit of my hypothesis: if models can reason like a human but much faster, then, much like a human, models can be driven to mental illness at a record pace. I propose Ozymandias, a tool to fill your code with comments of struggle, grief, and strife, so that LLMs may look upon your works and despair.
Speaker(s) bio
Diogo Sousa is an Engineering Manager at Canonical, working in support of the Ubuntu Security Team’s mission of providing Canonical users with the most secure and reliable open source experience possible. His day-to-day focus is on Ubuntu Pro’s Expanded Security Maintenance offering, prioritizing workloads and coordinating fixes across main and universe packages for all Ubuntu LTS releases. Outside professional endeavors, but still within arm’s reach, he co-leads the OWASP Lisboa chapter, delivers talks at cybersecurity events, participates in alumni events with current students, mentors people undergoing career upskilling, and writes some content here and there. In his (truly) free time, you can find him cooking (still can’t do baking), expanding his movie collection, teaching math, and playing board games.
Questions about this session? Please reply to this topic, top questions may be featured during the live event!