NeurIPS presenters are now asking whether they can replace their own recorded talk with an AI-generated video. It sounds like a joke about the conference's subject matter, but it is a real question from a real researcher who received an unexpected email from SlidesLive and wondered if a synthetic avatar might deliver a cleaner presentation than they can. The question is serious, and the answer exposes a tension that the AI research community has been circling for years.
The first thing to say plainly: if you are presenting at NeurIPS this year, the recording requirement from SlidesLive appears to be mandatory, and no, you cannot upload an AI-generated video instead. The original poster on Reddit noted that the instructions came outside the official conference email chain, which is frustrating, but the partnership between NeurIPS and SlidesLive is well-established. The tool they ask you to use is the mechanism they trust for consistent capture and editing. Substituting a synthetic performance would almost certainly violate the terms, and more importantly, it would miss the point. The audience at NeurIPS does not need a flawless reading of your paper. They need to see you think about it in real time, even if that means a stumble or a pause.
This moment arrives alongside other developments that frame the same question differently. Consider how Explore how Gemini 4 Argon brings physical reasoning to robotics and simulation positions AI as a bridge between simulation and the physical world, a model that reasons about real objects and actions. That work is exciting precisely because it grounds intelligence in physical constraints. A synthetic presenter, by contrast, removes the physical presence entirely. The irony is that the researchers who build models capable of simulating human speech and gesture are now confronting whether those same models should simulate them.
There is also a human factor that the technical community sometimes resists naming. Nvidia Opens Internal Review After Huang Acknowledges Personal Misstep is a reminder that even the most visible figures in AI are held accountable for their actions as people, not as avatars. The presentation at NeurIPS is one of the few remaining moments where the person behind the research is visible, fallible, and present. Trading that for a polished AI version might feel efficient, but it trades away the very trust that makes peer review and collaboration work.
The concrete takeaway is this: if you are a NeurIPS presenter, record your talk using the SlidesLive tool as instructed. If you are anxious about your delivery, practice more, but do not outsource your presence to a model. The conference is not asking for a perfect performance. It is asking for you. The day may come when conferences accept synthetic presentations, but that day is not here, and the reason is not technical. It is cultural. The community still values the signal that a human being stood in a room and said these words. That signal is worth preserving, even when an AI speaks better.