🚀💡 Quick take: Universal embeddings don’t scale well, and neural nets often miss the mark on quality—BM25 does it way better! We tossed Gemini all 46 docs and 1000 queries in one shot, and guess what? It nailed 100% of the queries in a single pass! 🤯 That’s a total win compared to top embedding models that struggle to hit even 60% recall@2. Oh, and by the way, LLMs are solid at retrieval too! Sometimes, the Language Model task is all you really need. 🧠✨ #DataScience #AI
🚀💡 Quick take: Universal embeddings don’t scale well, and neural nets often miss the mark on quality—BM25 does it way better!
We tossed Gemini all 46 docs and 1000 queries in one shot, and guess what? It nailed 100% of the queries in a single pass! 🤯 That’s a total win compared to top embedding models that struggle to hit even 60% recall@2.
Oh, and by the way, LLMs are solid at retrieval too! Sometimes, the Language Model task is all you really need. 🧠✨ #DataScience #AI
