Ideas
Bridging Seismic Imaging and Neural Fields
How far can NeRF-style models go for underground imaging when measurements are limited and noisy? I am interested in representations that respect underlying physics, not just black-box models, and in how priors from wave propagation can be folded into neural field architectures.
Curiosity Mechanisms in AI
Curiosity feels like one of the most human-looking parts of intelligence: noticing a gap, being pulled toward it, and exploring until the gap turns into a map. In RL we already have a small piece of this via intrinsic motivation — rewards for novelty, surprise, or learning progress — which help agents explore when external rewards are sparse.
But being curious about a topic is more than wandering in a maze. It needs memory of what is known, a model of uncertainty, and a planner that can choose actions like “read papers,” “run experiments,” or “compare baselines,” under constraints so that exploration has depth instead of becoming noise. If such systems tie curiosity to useful novelty — not just new, but better, simpler, or more general — they might realistically discover new algorithms rather than just random variations.
Billions Looking The Same: Why We Need to Build New Algorithms
I grew up in Turkiye, then went abroad for education — first the UK, then the US. Before I left, I expected cultural difference to feel louder: different humor, norms, and ways of seeing the world. Instead, what shocked me was how little 13,000 kilometers mattered. In London, in the US, back home, people reference the same viral moments, “main character” quotes, meme templates, and trending sounds.
As platforms optimize for engagement, they compress billions of different lives into a few dominant ways of seeing, thinking, and behaving. The system does not care about diversity of content; it cares about what works on most people most of the time. Feeds drift toward safe, familiar, easily digestible pieces that need no context or patience. Tools like duets, remixes, and templates lower the barrier to creation, but they also turn creation into filling in predefined blanks: variation without real divergence.
Sameness is not an accident; it is efficient. Platforms benefit from predictable content and responses. Advertisers benefit from aligned desires and aesthetics. Influencers who match the “algorithm-friendly” template are rewarded, sending a signal for others to imitate. Personalization quietly becomes a funnel: you start from “you,” but you end at “what performs.”
I do not think the only answer is “use social media less.” This is a systems problem. We need to build new recommendation algorithms that also optimize for cultural diversity, long-term curiosity, local context, and genuine novelty — systems that sometimes choose “interesting” over “safe,” reward original structures instead of only remixable formats, and keep the world from collapsing into one global aesthetic. Otherwise, “billions looking the same” will not just describe our content, but also who we become.