The best way to get the answer is not to ask questions, but to put the wrong answer. This is fun, but true indeed.
Google’s NotebookLLM doesn’t support some kind of website, like this https://bbycroft.net/llm
There is a tool for monitoring the browser via MPC named BrowserTools MCP. We can now debug the errors from the console directly from MCP clients (Claude desktop, Windsurf, etc).
HTTP/2 allows multiplexing, multiple requests can be processed simultaneously in a SINGLE TCP connection. I forgot this.
There is a technique to “log” asynchronously, I didn’t think about it before.
There is a mechanism to “block” and “unblock” read operations while there is an active write operation. This is more important when we have more replicas. I learned this from Clickhouse database design. In a generic case, let's say we have n replicas, we use m for read operations and n - m for write operations. This is how we design architecture to object to the P99. When writing the data, there is a built-in “signal” to tell all the replicas that there is an active writing operation. Also when the write is finished, there is a built-in mechanism to synchronize data to all replicas. The hardware is crucial here, but that’s the high-level flow of how they can handle an enormous number of records with very high SLA. Also, the columnar database is crucial here.
In software development, “greenfield” means starting from scratch, and “brownfield” means starting from an existing one.
Another AI code review tool: https://graphite.dev/docs/get-started

Today read

Til 2025-04-03

Subscribe to my newsletter

ShinaBR2

ShinaBR2