Nested Learning treats a single ML model not as one continuous process but as a system of interconnected, multi-level learning problems that are optimized simultaneously.

Nested Learning is trying to address the current implementations where knowledge is limited to the static information the LLM learned and the size of the context window.

And like Pokémon, there is a possibility that a neural network forgets old information as it learns something new, called catastrophic forgetting.

And in this case, its primary task is to look at its own code base, change it, benchmark the results, and continue this cycle. It works and is limited to the quality of the current coding agents, but even small incremental changes can be welcome.

When I got my Dreamcast for Christmas, it was snowing outside, which fitted perfectly with the time lapses in the game.

It was the first time I experienced a game like this, semi-open world, wandering around old 80s Japan.