The brain's memory is not a database; it consists of fixed weights that represent potentialities, guiding the generation of outputs based on inputs.

Cognition doesn't work like a retrieval system; when we try to remember something, we're not searching a space but generating from our past experiences.

I believe our brain doesn't actually retrieve information; instead, it fine-tunes and generates the next token autoregressively.

Global workspace theory suggests consciousness arises from a global workspace in our brains that processes inputs and broadcasts outputs, prompting us to consider how AI models might exhibit similar structures.

Memory consolidation happens through fine-tuning the weights of our neural network over time, rather than being a database of facts.

The best way to study and learn is not defined by the medium of material but by accessing memory systems that offset forgetting.

Focus involves activating circuits in the brain that analyze duration, path, and outcome of actions.

Even with large context windows in models, specialized tools are necessary for lookups to ensure accurate context management and memory selection.

I propose that cognition itself is autoregressive, similar to language, and that our brains have the machinery to support this process.