Qdrant reposted this
I'm a big science fiction fan, and for me, who started reading sci-fi books in the previous century, we are now crossing the border where those science fiction stories about the future become reality. Last week, I watched Yann LeCun's recent interview, where he discussed world models and where AI is heading, and it inspired me to think about the role of memory and information retrieval in this context. In parallel, completely independently, one of our engineers made a simple experiment with Qdrant and V-JEPA 2. What a coincidence! Or not... In any case, I've tried to put my thoughts into a content piece. Here it is: "Memories about the Future". Main theoretical thesis: If the next generation of AI thinks and learns in vectors, its memory should probably be a vector store. The most interesting move is what I call memories about the future: encode the present, roll it forward to imagine where things are heading, then retrieve the past experiences most relevant to that predicted state. Not "what's like now," but "what's like where this is going." That unlocks episodic memory, reusable skill libraries, and novelty detection for safety — all as similarity queries against a memory bank. We spent years building vector search to answer "what is". The work ahead is answering what could happen.