Friday, July 10, 202638 min read
transformers-from-scratch
Self-Attention by Hand: Queries, Keys, and Values
Lesson two builds a single attention head from a plain average of past tokens, deriving queries, keys, values, the 1/sqrt(d_k) scaling, and the causal mask in short, shape-annotated PyTorch.
positions
TSMC: The Leading Edge, and Who Pays to Build It
The world's logic foundry earns software margins on manufacturing, runs on one platform that is now 61 per cent of revenue, and is spending US$52-56bn this year on capacity its own customers underwrite.
the-divide
Weights in the Open: Should the Most Capable AI Models Be Released for Anyone to Download?
The open-model camp and the safety camp agree a frontier weight release cannot be recalled; they disagree about whether that is the feature or the fatal flaw.
the-brief
The Brief: The Cost War Goes Public
GPT-5.6 and Grok 4.5 reached the public within forty-eight hours, both racing on price and agentic coding — the clearest sign yet that the frontier's contest has moved from capability to cost.