9 chunks · 7 episodes
...ulty than writing for another person. People get bored and have a small working context window to synthesize thoughts within. LLMs never get bored and have millions of tokens of active context.
...hannel of spoken language. LLMs have a much larger internal "head" (weights and context window), and are able to translate on the fly better than any human in almost any domain.
As context windows get larger, the default personality of models matters incrementally less. Imagine that the model has a baseline perspective, and it takes tokens to ...
...s. LLMs can skim 100000x more than a human in that time frame. The limit is the context window, but it allows LLMs to read effectively "instantly". So of course LLMs will be great at the illusion of knowing you.
...fferent thing. You can lobotomize an agent by deleting the earlier parts of its context window. That doesn't feel like an individual, consistent entity.
...resented as if they were oracles. But of course they're actually LLM calls. The context window that it keeps appending to is what gives it a coherent throughline of agency. We can all see that context appending is not the final answer. You run ...
...uge deal. MCP felt cool, but it had a low ceiling and was easy to overwhelm the context window. Really what matters it's the ability of LLMs to do tool calling. More generally: to create software, to do things, whether in code, or with tool cal...
... what data it can see. That makes it so the reasoning doesn't muddy up the main context window, and vice versa. In complex adaptive systems, boundaries always emerge to handle the compounding cacophony. This reverse engineering of Claude Code b...
...even just focusing on the ones related to my job–was far too much to fit in the context window. So I added a feature to the Compendium allowing me to chat with any collection of cards. This feature is only enabled for me, so other viewers won't...