AI Papers of the Week
Every paper worth reading in AI, hand-picked one week at a time.
When to Retrieve During Reasoning
Most RAG systems retrieve once, before the model starts reasoning. Large reasoning models like o1 and R1 do not work that way. They generate 12k to 25k token chains of thought and hit knowledge gaps mid-inference, long after the retrieval window closed. ReaLM-Retrieve is a reasoning-aware retrieval framework that injects evidence during multi-step inference, detects uncertainty at reasoning-step granularity, and learns a policy for when external evidence actually helps. It achieves +10.1% absolute F1 over standard RAG across MuSiQue, HotpotQA, and 2WikiMultiHopQA, with 47% fewer retrieval calls than fixed-interval IRCoT, and hits 71.2% F1 on 2-4 hop MuSiQue with only 1.8 retrieval calls per question.
Co-evolving Decisions and Skills
Long-horizon agents fail in two ways: the decision-maker cannot decompose well, or the skill library goes stale. This paper introduces a co-evolution framework where an LLM decision agent and a dynamic skill bank improve each other through iterative refinement. The decision agent picks and chains skills, performance feedback updates both the policy and the skills, and new skills emerge by generalizing successful sequences instead of being hand-coded upfront. Most long-horizon agent stacks treat skills and decision-making as separate optimization problems, which is why they plateau. Co-evolution gives you adaptive planning and a growing library of reusable behaviors from a single loop, which is what you actually want when task structure is not predetermined: robotics, game agents, and complex planning.
DeepSeek V4
DeepSeek V4 is the first open model family built from the ground up around million-token contexts as a default rather than a bolt-on feature. The release includes DeepSeek-V4-Pro (1.6T total / 49B active) and DeepSeek-V4-Flash (284B total / 13B active), both trained natively at 1M context length. The tech report details a hybrid attention architecture, new training stability techniques, and a domain-specialist post-training pipeline that together push the open-source frontier much closer to GPT-5.2 and Gemini 3.0-Pro at a fraction of the cost.
Autogenesis
Static agents age quickly. As deployment environments change and new tools arrive, the agents that survive will be the ones that can safely rewrite themselves. This paper introduces Autogenesis, a self-evolving agent protocol where agents identify their own capability gaps, generate candidate improvements, validate them through testing, and integrate what works back into their own operational framework. No retraining and no human patching, just an ongoing loop of assessment, proposal, validation, and integration.
Attention to Mamba
Apple proposes a two-stage recipe for cross-architecture distillation from Transformers into Mamba. Naive distillation collapses teacher performance because a Mamba student cannot directly imitate softmax attention. The fix is to distill the transformer into a linearized-attention student using a kernel adaptation first, then transfer that student into a pure Mamba with no attention blocks. On a 1B model trained on 10B tokens, the Mamba student hits 14.11 perplexity against a 13.86 Pythia-1B teacher, nearly matching quality at linear-time inference cost.
Skill-RAG
Most RAG systems retrieve on every query, whether the model needs help or not. This is wasteful when the model already knows the answer and often too late when it does not. This paper introduces Skill-RAG, a failure-state-aware retrieval system that uses hidden-state probing to detect when an LLM is approaching a knowledge failure, then routes the query to a specialized retrieval strategy matched to the gap.
Self-Generated World Knowledge
How far are we from agents that can self-generate world knowledge? This paper proposes an outcome-based reward that measures how much an agent's self-generated world knowledge actually improves its task success rate, then trains with that signal and removes the external guidance at inference. The result is a 14B model that surpasses Gemini-2.5-Flash on web navigation and gains +20% on WebVoyager and WebWalker benchmarks.
Self-Evolving Logic Synthesis
EDA tools like ABC have been hand-tuned by humans for decades. NVIDIA shows they can evolve themselves. This work introduces the first self-evolving logic synthesis framework, a multi-agent LLM system that autonomously refines the entire ABC codebase, generates and tests candidate optimization sequences against standard benchmark circuits, then merges improvements back into the base tool. No human engineer in the loop.
Stateless Decision Memory
Most interesting AI agent papers right now are about capability. This one is about plumbing, and it is probably more important than it looks. Stateful agents do not scale horizontally. The moment you need thousands of concurrent agent instances running across containers, persistent per-agent state becomes the bottleneck. This paper proposes replacing active memory with immutable decision logs using event-sourcing principles from distributed systems.
There Will Be a Scientific Theory of Deep Learning
A position paper arguing that a genuine scientific theory of deep learning is already taking shape under the umbrella of "learning mechanics." The authors identify five converging research directions (solvable idealized models, tractable mathematical limits, simple macroscopic laws, hyperparameter theories, and universal cross-system behaviors) that share a common signature: they describe training dynamics, target coarse aggregate statistics, and commit to falsifiable quantitative predictions. The framing pushes back on skepticism about whether deep learning can have fundamental theory and positions learning mechanics as a complement to mechanistic interpretability, not a competitor.
MASS-RAG
Most real-world RAG failures come from retrieving technically-relevant but contextually useless documents, then forcing a single model to reconcile them. MASS-RAG is a multi-agent synthesis framework for retrieval-augmented generation where specialized agents handle distinct roles: retrieving candidate documents, assessing their actual relevance to the query, and synthesizing the final answer from evidence that actually contributes. Instead of one model doing everything, responsibility is decomposed across coordinated evaluators, which fits the direction the field is heading for deep research agents.
Diversity Collapse in Multi-Agent LLMs
Every multi-agent system pitch assumes agents explore different solutions, but this paper shows they converge on near-identical outputs over time, even across different architectures and different starting prompts. The authors call it diversity collapse. The cause is structural coupling: shared context, shared task descriptions, and mutual feedback pull every agent toward the same attractor. They measure it formally with metrics like the Vendi score, and the homogenization is real. The practical consequence is that multi-agent setups for brainstorming, hypothesis generation, and ideation only work if teams explicitly engineer isolated reasoning phases, decoupled evaluation, and heterogeneous starting conditions.
Automated Weak-to-Strong Researcher
Anthropic shows that Claude can run fully autonomous progress on scalable oversight research. A team of parallel Automated Alignment Researchers (AARs) built on Claude Opus 4.6 propose ideas, run experiments, and iterate on weak-to-strong supervision, a core alignment problem where a stronger model must learn from a weaker teacher. The system closes almost the entire remaining performance gap that human researchers could not, at a total cost of roughly $18K in tokens and model training.
AiScientist
Long-horizon AI research agents are mostly a state-management problem. Reasoning well for the next turn is not enough when ML research demands task setup, implementation, experiments, debugging, and evidence tracking over hours or days. This paper introduces AiScientist, a system for autonomous long-horizon engineering built around the principle of thin control and thick state. A top-level orchestrator manages stage-level progress while specialized agents repeatedly ground themselves in durable workspace artifacts.
AlphaEval
Agent evaluations are drifting away from production reality. Most benchmarks use clean tasks, well-specified requirements, deterministic metrics, and retrospective curation. Production work is messier, with implicit constraints, fragmented multimodal inputs, undeclared domain knowledge, long-horizon deliverables, and expert judgment that evolves over time. This paper introduces AlphaEval, a production-grounded benchmark evaluating agents as complete products rather than model APIs.
Nemotron 3 Super
NVIDIA introduces Nemotron 3 Super, an open 120B parameter model with 12B active parameters, built as a hybrid Mamba-Attention Mixture-of-Experts architecture optimized for agentic reasoning. The model targets long-context, high-throughput inference, a capability increasingly central to running agents reliably. It supports up to 1M context length while delivering up to 2.2x higher throughput than GPT-OSS-120B and 7.5x higher than Qwen3.5-122B, at comparable benchmark accuracy.
Memory Transfer Learning
Coding agents learn from experience, but that knowledge stays locked in silos. Solve a thousand SWE tasks, and none of that wisdom helps with competitive coding. This paper introduces Memory Transfer Learning, a framework where coding agents share a unified memory pool across six heterogeneous coding benchmarks, testing what transfers between domains and what does not.
Auto-Diagnose
Integration test failures are painful because the signal is buried in messy logs. Massive output, heterogeneous systems, low signal-to-noise ratio, and unclear root causes leave developers scrolling through thousands of lines. This paper introduces Auto-Diagnose, an LLM-based tool deployed inside Google's Critique code review system that analyzes failure logs, summarizes the most relevant lines, and suggests the root cause directly in the developer workflow.
Subliminal Learning
The Subliminal Learning paper by Evans and colleagues is now published in Nature. The work showed that LLMs can transmit traits (such as a preference for owls) through data that appears unrelated to that trait, like sequences of numbers that look meaningless on inspection. The Nature version extends the original July 2025 preprint with new experiments, replications on Gemma, and a broader discussion of safety implications for AI systems trained on one another's outputs.
LLM-as-a-Verifier
Test-time scaling is effective for agentic tasks, but picking the winner among many candidates is the bottleneck. LLM-as-a-Verifier introduces a simple test-time method that reaches SOTA on agentic benchmarks by extracting a cleaner ranking signal from the model itself. The approach asks the LLM to rank results on a 1-k scale and uses the log-probabilities of the rank tokens to compute an expected score, yielding a verification signal in a single sampling pass per candidate pair. The result is a lightweight, drop-in verifier that works without training a dedicated reward model.
WebXSkill
Web agents can navigate a page, but ask them to repeat a checkout flow they already completed and they start from scratch every time. WebXSkill is a skill learning framework where web agents extract reusable skills from synthetic trajectories, each pairing a parameterized action program with step-level natural language guidance. Two deployment modes let the agent either auto-execute skills as atomic tool calls (grounded) or follow them as step-by-step instructions while retaining autonomy to adapt (guided). On WebArena, WebXSkill improves task success by up to 9.8 points over baselines. On WebVoyager, grounded mode reaches 86.1%, a 14.2-point gain, and skills even transfer across environments.
Muses-Bench
Every agent framework assumes one user giving instructions, but in real team workflows agents have multiple bosses with conflicting goals, private information, and different authority levels. Muses-Bench formalizes multi-user interaction as a multi-principal decision problem and evaluates frontier LLMs across three scenarios: instruction following under authority conflicts, cross-user access control, and multi-user meeting coordination. Gemini-3-Pro tops the leaderboard at just 85.6% average, and no model exceeds 64.8% on meeting coordination. Privacy-utility tradeoffs are brutal: Grok-3-Mini scores 99.6% on privacy but collapses to 60.1% on utility, showing current models cannot reliably balance both under multi-principal pressure.
Neural Computers
Researchers from Meta AI and KAUST propose Neural Computers (NCs), an emerging machine form that unifies computation, memory, and I/O in a single learned runtime state. Unlike conventional computers that execute explicit programs, agents that act over external environments, or world models that learn dynamics, NCs aim to make the model itself the running computer, establishing a new computing paradigm.
Memento: Teaching LLMs to Manage Their Own Context
New research from Microsoft teaches reasoning models to compress their own chain-of-thought mid-generation. Memento trains models to segment reasoning into blocks, summarize each block into a compact “memento,” and then evict the original block from the KV cache. The model continues reasoning from mementos alone, cutting peak memory by 2-3x while nearly doubling throughput.