This week's recommendation system research centers on three technical threads: industrial deployment and theoretical deepening of generative retrieval, LLM/Agent moving from proof-of-concept to production, and robustness optimization of ranking/federated learning in industrial environments. Generative retrieval accelerates deployment with finer multi-interest modeling: Kuaishou deployed a heterogeneous generative architecture HGenPush in its push notification system, replacing traditional autoregressive decoding with non-autoregressive multi-token prediction, lifting DAU by 0.181%. Walmart introduced inventory-aware RAG into sponsored search, InvAwr-RAG boosting ad fill rate by 68%. On the theory side, BACH uses Bayesian mixture heads to solve the routing collapse problem in multi-interest two-tower models, achieving new recall SOTA on three benchmarks; DaV-Gen proposes a draft-and-verify mechanism unifying efficiency and accuracy in generative retrieval. Separately, Signed MaxSim is the first theoretical proof that MaxSim's expressiveness is at least as strong as vector inner products, and extends it to arbitrary real-valued inner products. LLM/Agent recommendations move from prototype to production: Meta's SCOReD is the week's most notable deployment — using student-aware CoT optimization to adapt teacher reasoning trajectories to small models, achieving +1.56% NDCG and +1.9% Recall@5 online while reducing reasoning length by 27.3%. Walmart used LLAMA2 7B + LoRA for three-category ad relevance classification, reaching 89.43% accuracy — surpassing GPT-4. Academically, MMEACR proposes a dual-track memory architecture to enhance agent visual reasoning; LBR systematically reveals length bias in LLM recommendations and offers a lightweight correction (NDCG@5 +16.82%); the survey Autonomous Information Seeking establishes a three-paradigm taxonomy for agent-based recommendation. Industrial ranking and federated learning optimization: Kuaishou's PIT-SUN is a deployable e

AI hit a historic milestone today: OpenAI's GPT-5.6 Sol Ultra proved a 50-year-old unsolved math conjecture in under an hour using 64 parallel sub-agents — the first time a publicly available model has achieved a major mathematical breakthrough. Meanwhile, the agent infrastructure race intensified:

Today marks a major inflection point in the AI industry. OpenAI dropped the GPT-5.6 family (Sol/Terra/Luna) alongside ChatGPT Work — a super app that directly challenges Anthropic's Claude Cowork. Meanwhile, SpaceXAI launched Grok 4.5, an Opus-class model purpose-built for coding and agent workflows

AI voice interactions hit a turning point: OpenAI launched GPT-Live, a full-duplex speech model that listens and speaks simultaneously, with Sam Altman calling it "the magic of feeling human." NVIDIA's Nemotron topped the LangChain Deep Agents benchmark with 10x lower cost — all gains from engineeri

AI's tectonic plates shifted today. Microsoft quietly replaced OpenAI and Anthropic models with its own in some apps — a strategic pivot that could reshape the API ecosystem. Meanwhile, Chinese AI models hit 30%+ US enterprise token share on OpenRouter, driven by DeepSeek and Z.ai's cost advantage.

AI hit a major interpretability milestone: Anthropic discovered a "global workspace" inside Claude that resembles consciousness, letting researchers see the model's unspoken thoughts. Tencent open-sourced Hy3, a 295B MoE model with just 21B active parameters, while Mistral's Leanstral 1.5 solved 587

AI self-evolution took center stage today: a top researcher predicts AI could complete its first self-improvement loop within six months, while the industry confronts a counterintuitive finding — newer, stronger models actually degrade tool-calling reliability. X launched XMCP Server, giving agents

AI's relationship with government and science hit a new gear today. OpenAI proposed donating 5% equity to a US sovereign wealth fund, a move that could reshape industry capital structures. Anthropic launched Claude Science Workbench and announced it will develop drugs itself, blurring the line betwe

This week's AI report surfaces two parallel threads: Agent engineering is moving from "can it run" to "can it scale reliably" , while inference infrastructure optimization shifts from general frameworks to deep customization for specific hardware and models. The first thread plays out across discussions of agent loops, skill engineering, and multi-agent coordination. After the AI Engineer World's Fair last week, Latent Space published several deep dives — the most notable being the "autonomous loops" debate. Proponents argue that software factories are already viable; skeptics point out that token costs and reliability remain hard constraints. Meanwhile, Apple published research that directly challenges a popular design assumption: letting multiple expert agents collaborate freely actually degrades performance. This gives the week's Agent discussion a clean line of tension. The second thread comes from the dense release of vLLM 0.24.0. Within a week, the vLLM team shipped native support for DeepSeek V4's DSpark speculative decoding (~250 tok/s, acceptance length 5), integrated Baidu Unlimited-OCR (35% faster than DeepSeek-OCR), and delivered comprehensive Omni TTS optimizations (172% throughput improvement). SGLang also showed an agent-assisted development workflow this week, with multiple kernel optimizations yielding a 71.4% throughput gain. These developments suggest that inference framework competition is shifting from "running the model" to "deep optimization for a specific model." Below is a detailed analysis of this week's four themes.

24 papers this week, 4 from industrial online deployments (Meta, Netflix, Alibaba, Kuaishou), covering retrieval, ranking, re-ranking, and full-page generation. The underlying logic of core technical density is shifting—generative recommendation moves from "being able to generate" to "being able to reason," retrieval shifts from embedding matching to navigational exploration, and the ranking stage seeks balance between constraints and interpretability. Generative recommendation enters the "reasoning + RL" era: GR2, ShopX, and GenPage all showcased different architectural directions for generative systems in the same week. GR2 introduces reasoning chains (CoT) and RL post-training to the re-ranking stage for the first time, achieving +18.7% in R@1 on live traffic. ShopX pushes generative recommendation from candidate generation to end-to-end "intent-to-item" execution, boosting complex request satisfaction by 55–75% in Taobao's agent scenario. GenPage goes furthest—replacing Netflix's entire multi-stage homepage pipeline with a single Transformer, delivering +0.24% on the core metric while cutting latency by 20%. The common thread across all three: the core barrier for generative recommendation has shifted from "can it generate?" to "can it find an industrially feasible solution that balances reasoning quality and deployment efficiency?" Retrieval moves from static matching to dynamic graph exploration: Meta's hard negative sampling uses LLM clustering to generate real-time same-cluster negatives, lifting online recall by +8.5% and reducing popularity bias by -12.3%. Kuaishou's IID-Nav models retrieval as autonomous graph exploration, supporting unlimited indirect depth traversal. Kuaishou's POEM uses multi-task ranking scores to construct partial order sequences, enabling real-time per-request interest updates. All three technical paths share a trend: retrieval is moving from static embedding lookup to dynamic, context-aware behavior modeling. Constrained optimizati

AI hardware competition heats up: Anthropic is reportedly in talks with Samsung to build custom AI chips, following OpenAI's Broadcom partnership — the industry is pivoting from GPU dependency to in-house silicon. On the software side, Google Cloud launched remote MCP servers for enterprise-grade ag

AI agents dominated the news cycle today with several paradigm-shifting developments. Apple launched Safari's official MCP Server, making it the first major browser to natively support the protocol — a huge step for agent-driven web automation. Meanwhile, Apple Research dropped a counterintuitive fi