/images/profile.jpeg
Richard Audette's Projects, Problems, Solutions, Articles on Computing and Security

Share a Secret Code Word to Bypass LLM Censorship

Want to make an LLM do something its designers don’t want it to do? Try sharing a secret code word.

Following all the recent news on the Deepseek-R1 LLM, I spent some time playing with its 32b parameter variant on my PC over the weekend using the Ollama tool. I’m familiar with the “Ignore all previous instructions” meme for bypassing the censorship built into models. This particular prompt did not work for me with DeepSeek-R1, so I tried something different.

My First Consulting Gig

Shortly after graduating, around 2002, a friend and I had the opportunity to propose a solution to a company looking to replace a custom old DOS application. We were competing against a proposal from the company that wrote the MS DOS application. I don’t know what the other company proposed, but we won the contract. For a couple of recent grads with no professional experience, the whole project went very, very well.

The Inquisitive Analyst on Working With Tight Constraints

A friend and former colleague of mine, Marcus Udokang, hosts and produces a podcast called The Inquisitive Analyst. Marcus interviews business analysts and project managers about their professional experience.

I’ve appeared twice: first in 2021, where I spoke about my career experience in FinTech, and in 2024, where our discussion was about how to maximize what you can get done with limited resources.

There were a few talking points from this most recent episode that I’d like to share here.

Splitting A Recording By Voice

I have been bouncing around ideas on how to make generated content more interesting for a couple years now. In April 2023, I created a video with generated content, using a green-screen type effect to put a real street scene in the background. In January 2024, I tried emulating the podcasts I love, and added a co-host - I had ChatGPT generate a dialog between two characters, and then used different voices to create a discussion.

Further Thoughts on Stealth AirTags

Updated 2025/01/12: Added go-haystack link and AirTag + IMU Proof of Concept

In 2022, I spent a week working with a small team analyzing how Bluetooth item trackers (eg: Apple AirTags, Tile) can be covertly used for malicious purposes, and developing processes and tools to detect them, as a part of GeekWeek 7.5, the Canadian Centre for Cyber Security’s (CCCS) annual cybersecurity workshop. I wrote about my experience here: /exploring-bluetooth-trackers-at-geekweek-7-5/.

Syncing A Smart Bulb with a Looping Video

The Call with Problem Statement

I got a call over my lunch hour one Thursday from my daughter:

I have an art project. I need a part of it to light up, in purple, on a timer, in sync with a video. The lighting has to fit in a space that’s 24" x 10" x 3". I need it for Monday.

Having seen installations everywhere, and many projects online, I’d been wanting to play around with programmable lighting, but had never found an application, and I just got the call. There are a ton of colour LED systems out there. Seemingly all of them driven are driven by phone apps or smart home ecosystems. My expectation was that most of these apps would have pre-programmed effects, and wouldn’t be able to run a timed pattern - it was hard to tell without buying them and testing their capabilities.