AI basics
What does MCP actually do?
It gives AI apps a shared way to connect to tools and information outside the chat. The useful question is what you’re connecting—and what you’re letting it do.
Meet Mochi · @kawaiistbets
AI Jargon
in little bites
AI, chips, and earnings, explained in about a minute. A little less jargon. A little more “oh, I get it.”
Free to explore. Original sources included.
wait, what does thatactually mean?your curious little guide ♡
One useful idea at a timeSources you can openResearch from tkal
Start with this one
Your AI got a common plug.
Mochi’s Minute · MCP
The little learning shelf
What are you curious about?
Start anywhere. No homework required.
What does MCP actually connect?
A common plug for AI apps, tools, and information.
Read Mochi’s story → What does MCP actually connect?
Why does a long chat lose the thread?
Meet the context window: the working space behind a response.
Read Mochi’s story → Why does a long chat lose the thread?
What does your AI look up first?
Meet RAG: the AI checks outside documents before it answers.
Read Mochi’s story → What does your AI look up first?
What happens when AI answers?
Inference is a trained model doing work on a new input.
Read Mochi’s story → What happens when AI answers?
Why put memory next to an AI chip?
A fast chip needs a fast way to get its data.
Read Mochi’s story → Why put memory next to an AI chip?
What is your chatbot counting?
Tokens are the pieces of text a model reads and writes.
Read Mochi’s story → What is your chatbot counting?
How much data center power never reaches a chip?
Meet PUE: overhead power compared with computing power.
Read Mochi’s story → How much data center power never reaches a chip?Why would a chip skip precision on purpose?
An NPU, a neural processing unit, trades exact math for AI work on less power.
Read Mochi’s story → Why would a chip skip precision on purpose?Why would Google build a chip for one job?
A TPU, a tensor processing unit, is Google’s custom chip for large matrix math.
Read Mochi’s story → Why would Google build a chip for one job?Why would an AI server borrow your phone’s memory?
SOCAMM puts low-power mobile memory on a small module for data center servers.
Read Mochi’s story → Why would an AI server borrow your phone’s memory?Why would a data center swap plugs for light?
Photonics moves data with light pulses, and Nvidia wants it right on the chip’s package.
Read Mochi’s story → Why would a data center swap plugs for light?Why did a gamer chip end up running AI?
A GPU does the same math on lots of data at once, and AI workloads lean on it.
Read Mochi’s story → Why did a gamer chip end up running AI?Will this AI model fit on your GPU?
VRAM is the GPU’s own fast memory, and a model’s weights have to fit inside it.
Read Mochi’s story → Will this AI model fit on your GPU?Who pays for the cloud build?
Alphabet’s Q3 2026 shows why cash flow deserves its own page.
Read Mochi’s story → Who pays for the cloud build?
More revenue. More stuff sold?
A bigger sales number can have more than one ingredient.
Read Mochi’s story → More revenue. More stuff sold?No matching stories yet. Try another word or choose Everything.