Can a single model generate all modalities without external encoders?
Most multimodal systems rely on separate encoders for each modality. This research explores whether training a unified foundation model on discrete tokens across text, image, video, and speech can enable any-to-any generation without those external components.
Why do unified image generators fail on non-Latin scripts? Does multimodal zero-shot performance actually generalize or interpolate? Can we solve modality competition through architectural design? How do language models learn to think like humans? Are text-only language models fundamentally limited by abstraction?
Unified image generators inherit training data bias visible as broken script rendering Multimodal zero-shot performance is driven by concept frequency in pretraining data Modality competition is architectural, not inherent to multimodal training Language models develop human-like cognitive biases and reasoning patterns Text-only LLMs inherit the abstraction limits embedded in language itself
Do discrete tokenized modalities preserve information better than continuous embeddings? What emergent abilities appear only in truly unified multimodal systems? How does causal multimodal modeling differ from encoder-decoder architectures? Can this whole-artifact principle apply to other generative tasks?