What’s of interest? I Built a Scam Detector for My Family Using AI Running on My Laptop | Ian O’Byrne
Tell me more!
Back in 2023, I worked with a team on a project called TrustSense. An AI tool to help people figure out whether an email, social post, or article was trying to manipulate or deceive them. The project emerged from Carnegie Mellon’s Generative AI Innovation Incubator hackathon. This was a three-track series organized through CMU’s School of Computer Science, focused on Medicine and Health, Education and the Future of Work, and Finance and Economics.
I worked across multiple tracks. In the Finance and Economics track, our team built TrustSense and reached the final round. I documented the full build across five posts as we went: the introduction, the steps involved in building it, the quest for trustworthy training data, how we thought about user privacy, and how we tested different AI models.
That last post is worth mentioning here because it’s directly relevant to what changed between then and now. In 2023, we tried Llama 2, and it hallucinated constantly, lacked explainability, and we couldn’t reliably control its output. We tried GPT-3.5 but ran into token limits during the hackathon. We ended up building a custom LSTM (Long Short-Term Memory) model with a LIME (Local Interpretable Model-agnostic Explanations) explainer on top.
In plain English, this means that it could be trained specifically to read emails and spot patterns. On top of that, we added a tool that could show its work. Instead of just saying “this email is suspicious,” it could highlight the specific words and phrases that triggered the warning. We didn’t want a black box that gave you an answer you had to trust blindly. We wanted something that could point at the email and say: “It’s this line here, and this one.”
A few weeks ago, I rebuilt it. Alone. Over a weekend. Using AI running entirely on my laptop.
I’m not saying that to sound impressive. I’m saying it because the gap between what that took in 2023 and what it takes now is kind of staggering, and I think it matters for anyone trying to stay on top of what AI can actually do. Not the hype, but the practical reality.
Where is it?: I Built a Scam Detector for My Family Using AI Running on My Laptop | Ian O'Byrne
This is one among many items I will regularly tag in Pinboard as oegconnect, and automatically post tagged as #OEGConnect to Mastodon. Do you know of something else we should share like this? Just reply below and we will check it out.
Or share it directly to the OEG Connect Sharing Zone