AI Circle x Contra: Frontier Creative AI - Model Signal through Divergence
Across creative workflows like design, copywriting, and video editing, model outputs sit somewhere between stereotypical - the most expected, recognizable, on-distribution answer — and surprising - novel, unexpected, sometimes wildly off-distribution. Both can be the right output. The same surprising output is failure in one workflow and the source of creativity in another.
The question isn’t how to control models, but how to harness divergence as a feature for creativity, not a failure of quality output.
This panel brings together builders, researchers, and practitioners at the edge of creative AI to unpack the following:
Where and why models diverge in real-world creative tasks
The tension between alignment, determinism, and generative exploration
How teams are designing systems, prompts, and feedback loops to harness creative variance
What good looks like when outputs aren’t easily benchmarkable
The role of human taste, evaluation, and iteration in shaping creative systems
We’ll explore how divergence shows up across modalities, and what it reveals about the next generation of model behavior.
Because the frontier isn’t just smarter models.
It’s models that delightfully surprise us, and knowing what to do when they do.
Panelists
Mariya Vasileva: Sr. Research Scientist at Meta SuperIntelligence Labs
Aakanksha: Member of Technical Staff, Post-training at Cohere
Aspen Hopkins: ML PhD w/background in HCI / Data Visualization from MIT CSAIL
Priyaa Kalyanaraman: CEO & Cofounder at Lica.World
Agenda
5:30 - 6:30pm: Arrival + refreshments + networking
6:30 - 7:30pm: Panel discussion
7:30 - 9:00pm: Open discussion + drinks
Who This Is For
Builders working on creative AI products (text, image, video, multimodal)
Researchers exploring alignment, RL, and generative behavior
Designers, engineers, and operators shaping AI-assisted creative workflows
Anyone thinking deeply about what “good” looks like when outputs aren’t deterministic
About the hosts:
Contra Labs harnesses the power of their 1.5M+ creative network to train SOTA models. Learn more about Contra Labs.
AI Circle curates a community of practitioners developing and deploying AI systems across research, startups, and enterprise. Learn more about AI Circle.
Limited capacity.
Apply to attend.