X's algorithm soon to be purely AI, configurable by asking Grok (twitter.com)

X announced it will replace its current ranking stack with a “purely AI” algorithm that users can directly configure by asking Grok, the platform’s conversational AI. Instead of static, hand-tuned ranking rules, Grok will accept natural-language instructions (e.g., “show me more science and less politics”) and adjust feed signals and ranking behavior in real time. The move reframes the recommender as a controllable LLM-driven layer that translates user intents into personalization parameters, rather than a fixed codebase of heuristics. For the AI/ML community this is both an experiment and a challenge: it accelerates research into natural-language control of recommender systems, real-time policy translation, and safe human-in-the-loop personalization. Key technical implications include mapping free-form user prompts to ranking weights, scaling low-latency inference across billions of timelines, maintaining robustness to adversarial prompts, and auditing for fairness and misinformation. It also raises evaluation questions—short-term engagement vs. user well-being—and operational concerns like logging, privacy, and model update governance. Researchers and engineers will watch how X handles interpretability, guardrails, and long-term feedback loops, since success or failure here could reshape how social platforms let AI mediate and be steered by users.