Image created with gemini-3.1-flash-image-preview with claude-opus-4.7. Image prompt: Using the provided reference images, keep the authentic Sonoran Desert trail setting, brown wooden post, off-white sign face, and ranger-style bold typography, but replace the header with ‘RAG’ in bold all-caps and add fictional trail entries like ‘→ Source Documents 0.4 mi’, ‘← Context Window 1.2 mi’, ‘→ Citation Spring 2.7 mi’; lean a weathered wooden card-catalog drawer with visible index cards against the sign post, and place a small cactus wren on top holding a single index card in its beak, with saguaros, volcanic rock, and the hazy valley vista behind, photorealistic midday Arizona light.
Retrieval is all about signal, architecture, not raw power. 🦾The late interaction LateOn 0.1B competing with the dense Gemini 2
https://x.com/LightOnIO/status/2052095548098822477
// When to Retrieve During Reasoning // Pay attention to this one, AI devs. (bookmark it) Most RAG systems retrieve once, before the model starts reasoning. Large reasoning models like o1 and R1 don’t work that way. They generate 12k-25k token chains of thought and hit
https://x.com/omarsar0/status/2049954716298494386
Leave a Reply