Wikipedia link race: choosing among thousands of options
The model chooses the next step from hundreds or thousands of links, demonstrating fast decisions with constrained outputs. The post includes a 4:04 video.
Evaluating RAG Reranking with Jev
Shares insights on using Jev to rerank RAG candidates, covering latency, cost, and failure points, with practical tips like setting thresholds and freezing candidates.
RAGのリランキングを Jev で評価した。 ベクトル検索の上位候補を並べ直し、順位改善を測る設計ができる。 遅延・費用・障害点を踏まえ、導入価値を判断するのがいいかも。 ゲートを先に決める／正解を引用文で持つ／候補を凍結する、実践のヒントなのがいい。
· 0 likesOpen on X
More like this
Semantic Querying Use Case with Jev and TypeSafe AI
Nathan LeClaire shares a semantic querying use case for Jev and TypeSafe AI, noting people have only scratched the surface.
Building a live BS meter for the Trump vs Kamala debate with Jev
The author used Jev to run a live BS meter on every sentence of the Trump vs Kamala debate, with 5 yes/no questions per candidate, totaling 1,191 Jev calls, 1.18M tokens, 415 ms median latency, and $0.0497 total cost.
