Scaling Your Creative Output with AI: Lessons from SANS Holiday Hack

The talk opens with a plain explanation of generative AI as a way of compressing language, images and audio into a compact representation and producing new output from it. It then surveys the tool landscape, from hosted commercial services to free open-source models and desktop applications that let people run models privately on their own hardware. Open models often ship without the safety guardrails found in commercial services, which the speaker flags as a topic worth discussing.

The middle section argues that a real project with a clear purpose is the best starting point, and that unexpected output can seed better ideas. Examples include a small web interface for comparing prompt variants and workflows for matching map tiles to photos. The main case study comes from the SANS Holiday Hack Challenge, where the design team used a language model to draft NPC dialogue from years of earlier lines, image models to produce map tiles and a set of whimsical fish artwork, and a text-to-speech pipeline to voice dialogue stored in a spreadsheet-driven game engine. The speaker estimated the dialogue tool roughly halved writing time, and more artists were able to contribute to the game's look.

The closing looks at the risks: lawsuits over training data are reshaping commercial tools, low-effort content will flood platforms, and cloned voices and images erode trust in media. The main takeaway is that communicating ideas is getting easier, so practitioners who pair AI with solid data and real projects should gain an edge.