Deep fake it till you make it!

The talk opens with a light icebreaker in which the speaker swaps a face in a photo using AI tools, then moves into the history of machine intelligence. It covers a 1984 prediction about synthetic life, Vernor Vinge's idea of the technological singularity, and Nick Bostrom's thinking on risky technologies and a future where AI solves every problem. The core argument is that easy deepfake video and voice cloning together create a dangerous combination.

The middle of the session is a series of live demonstrations using locally run open-source language models and image generators. The speaker builds a fictional tabletop-game character with a matching dating profile, generates portrait images for it, and shows how a single photo can be used to place a face into new poses. He then clones voices from public audio to produce a mock birthday message in the voice of a famous action star and a well-known political figure. With consent, he also clones a volunteer's voice from online recordings to deliver a short message. Some image generation steps did not work well on stage, and the speaker presents this as typical of live demos.

The closing takeaway is that the technology is both exciting and worrying. Listeners should not trust everything they see or hear online, and should pause and verify before acting on unexpected requests, especially ones that appear to come from people they know. Because a few minutes of public video or audio can be enough to build a convincing clone, people should think carefully about what they publish. The speaker ends by reflecting on how one's point of view determines whether the world feels like a utopia or a dystopia.