
Designing Speech Transcription for Context, Not Just Latency
Why full audio transcription outperforms streaming transcription, and how encoder-decoder models, cross-attention, and faster inference shape product tradeoffs.
Johnson Shi's thoughts on technology, markets, and the systems shaping our future.
Live Last read · last published · next check

Why full audio transcription outperforms streaming transcription, and how encoder-decoder models, cross-attention, and faster inference shape product tradeoffs.

DR-DCI is an optimization built on top of RAG for letting agents run precise, verifiable search across large document collections without scanning the whole corpus on every query.

Short depreciation cycles, long depreciation cycles, and why neoclouds, AI providers, and chipmakers all win either way.

The bottleneck always moves — why capacity engineering looks more like managing a Costco than setting signals once and walking away.

A structural pattern has reshaped servers, labor, and software licenses. AI is now running the same playbook — simultaneously — on software production and human capital.

Why the smartest way to teach an AI agent is a Markdown file and a YAML header

How typed function signatures replace prompt engineering for scalable AI products.

Is your organization looking for AI that thinks with you or for you? Do you need AI that does things with you or does things for you?

Exa’s bet that owning agentic web I/O can be as powerful as Stripe owning payment rails—potentially taxing agentic information streams the way Stripe taxes transactions.

Open source is part idealism, part strategy for the IT industry. Like punch-card programming standards, it powered computing and built entire industries with room for profit.