attention runs the world. no platform comes close to X on density. the best founders, investors and engineers alive are reading the same timeline. pay attention to where people pay attention, then build there.

ATTENTION BURLINGTON, ONTARIO The mother of a convicted lSlS t*rrorist is running for school board for the Halton School District. Khdiga Metwally’s son is currently serving a 40-year sentence for plotting to blow up NYC. Do you want this woman in charge of your kids’ education?

ATTENTION STOCK TRADERS > this is a heads up. This week, I’ll be announcing a FREE LIVE, IN-PERSON WORKSHOP with me and a few key members of my team. This will be an all-day event where we’ll teach you how to find and manage winning stocks using the Minervini Markets 360° platform and the same time-tested principles and processes I use every day. A registration link will be provided when we officially announce the event with location, date & time. The event date is only a few weeks away. Seating is extremely limited, and we expect it to fill up quickly. When registration opens, don’t wait—act fast and reserve your seat.

ATTENTION IS NOW SHIFTING TO U.S. INFLATION DATA LATER THIS WEEK, WHICH COULD DETERMINE WHETHER THE FED HIKES AT ITS SEPT. 15-16 MEETING, AS STRONG JOBS DATA HAS ALREADY LIFTED HIKE EXPECTATIONS AND A HOTTER READING COULD BOOST THE DOLLAR.

[Attention] Attention is how one token pulls relevant information from itself and the tokens before it. At each layer, learned projections turn each token’s current representation into three vectors: a query, a key, and a value. The query represents what the current position is looking for. Keys describe what each token can be matched on, while values carry the information that can be pulled in. For “runs,” one attention head compares its query with the keys of “The,” “chip,” “Alice,” “designed,” and “runs” itself. Each comparison is a scaled dot product, q · k / √d. Softmax turns the scores into weights that add up to one, and the values are multiplied by those weights and summed. A transformer runs several heads in parallel. Each has its own learned projections, so it can view the same tokens differently. Their outputs are combined and passed through the remaining layers. A head may emphasize the subject while another captures a different relationship, but these are learned tendencies, not roles assigned in advance. But where did all those keys and values come from? And when the next token arrives, does the model have to build the earlier ones all over again? It does not. Tomorrow: KV cache. The paper that introduced the transformer: Vaswani et al., NeurIPS 2017.