Categories: Omni-Centralism

20 entries

You want one natively omni-modal, end-to-end model that sees, hears, speaks and draws as one.

Visual-Token Statism
Pixels should be discretized into visual tokens so images and video can be modeled with the same sequence machinery as language.
Bitter-Lesson Orthodoxy
General methods that scale with computation, namely search and learning, always beat clever human-designed priors in the long run.
Diffusion-Autoregressive Syncretism
Autoregression and diffusion are complementary generative principles and belong together in one model, each used where it fits best.
Omni-Model Absolutism
The future belongs to a single natively omni-modal model that sees, hears and speaks in real time; everything else is transitional.
Any-to-Any Imperialism
One model should take any modality as input and produce any modality as output, annexing every specialist system along the way.
VLA Monarchism
A single vision-language-action model, built on a pretrained vision-language backbone, should be the sovereign brain of every robot.
Unified-Vocabulary Centralism
Images, audio and actions should be discretized into one shared vocabulary so a single sequence model can handle all of them.
Linguistic Multimodalism
Aims at a fully multimodal model that understands and generates every modality, using language modeling as the engine to get there.
Native-Fusion Centralism
Modalities must be fused from the very first layer and the very first token of pretraining, never grafted on afterward.
World-Simulator Statism
Large video generation models, trained at scale, are learning to simulate the physical world and will become general world simulators.
Scaling-Law Planism
Scaling laws make AI progress forecastable, so model size, data and compute should be allocated by plan from fitted curves.
Interleaved Statism
Web documents that interleave images and text are the new textbooks, and multimodal models should be pretrained on them at scale.
Early-Fusion Statism
Modalities should be merged at the input layer and processed together by one network, rather than combined late from separate encoders.
Audio-Native Doctrine
Holds that speech and sound should be modeled natively, end to end, rather than converted to text and back.
Pixel Totalitarianism
Text is just another image, so models should read rendered pixels instead of tokens and absorb language into vision.
Chat Absolutism
The chat box is the one true interface to AI, and every capability should be delivered as a turn in a conversation.
Long-Context Expansionism
Believes ever-longer context windows will replace retrieval systems and external memory, letting the model read everything at once.
Byte-Level Populism
Tokenizers are an unelected elite; models should read raw bytes and learn their own units of meaning.
Curiosity Liberalism
Agents should be driven by intrinsic motivation, rewarding novelty and learning progress, rather than waiting for external rewards.
Representation Centralism
Holds that one general-purpose learned representation, trained once at scale, should serve every downstream task.