I noted that the 512d binary quantized embedding outperform 3096d float32 embeddings from OpenAI v3 large.
- Greedy strategy
I explained that RULER, implemented in OpenPipe ART, lets the engineer write the evaluation criteria in plain English.
I highlighted that Opik already implements this observability infrastructure for LLM apps and is open source.
These notes complement my earlier September 2026 LinkedIn series on RAG architecture, decoding strategies, reward engineering, and senior‑engineer heuristics. See linkedin september 2026 posts for the full recap.
- Multinomial sampling strategy
- Beam search tries to approximate the true global maximization
- Contrastive search