PulseAugur
LIVE 12:24:51
research · [1 source] ·
0
research

Qdrant warns users against trusting its new BM42 model

Qdrant has released a new open-source embedding model called BM42, but the company is urging users not to trust its performance. The model was trained on a dataset that Qdrant itself curated, and the company acknowledges potential biases and limitations. They recommend thorough evaluation before deployment in production environments. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

RANK_REASON Release of an open-source model with a caveat about its trustworthiness.

Read on Smol AINews →

COVERAGE [1]

  1. Smol AINews TIER_1 ·

    Qdrant's BM42: "Please don't trust us"

    **Qdrant** attempted to replace BM25 and SPLADE with a new method called "BM42" combining transformer attention and collection-wide statistics for semantic and keyword search, but their evaluation using the Quora dataset was flawed. **Nils Reimers** from **Cohere** reran BM42 on …