Live intelligence

Latest AI updates

571 updates

Research

Microsoft Research Focus #97: Agent Frameworks, Coding-Agent Limits, and BioEmu2 Preview

Microsoft Research's latest Research Focus edition rounds up new work on self-improving agents (ForkPilot), coding-agent evaluation (CABRA), and skill learning from activity logs (TeleTune). It also highlights a study of Copilot use across industries, Shiksha Copilot in Indian classrooms, a probabilistic bias-correction framework for extended-range weather forecasting, and the public preview of biomolecular foundation model BioEmu2. The episode points to concrete research directions in agent reliability, workplace AI adoption, and AI for science.

Microsoft Research6d agoScore 68

Microsoft ResearchAI agentsAI for science

Research Editor selected

Infini-AI-Lab Releases Hermit, an Open-Source Harness Adapter Framework for AI Agents

Infini-AI-Lab has released Hermit v0.1.0, an Apache-2.0 Python framework that trains an external "harness adapter" while the underlying agent model and harness remain frozen, so Codex or Claude Code agents can be adapted to new tasks without fine-tuning or forking. Hermit uses an RL-style loop with a Trainer, a judge and an Evolver, ships five evolution methods (GEPA, SkillOpt, MWM, Harness², Continual Harness) and versions every adapter in a git tree. Self-reported results with GLM 5.3 Flash include MWM reaching 0.795 on CL-Bench database exploration versus a 0.580 in-context baseline, and Harness² improving OfficeQA to 0.651 from 0.541 with no adapter. Adapters can be deployed to Codex, Claude Code or Pi, or exported as native plugins.

Why it matters: It gives agent developers a concrete, installable open-source way to train and deploy portable adapter layers for existing coding agents, with benchmark numbers and clear caveats.

MarkTechPostOct 29, 2026Score 84

Infini-AI-Labagent optimizationharness adapters

Research

ReForest Combines Isolation Forest With LLM Reasoning for Anomaly Detection

ReForest is a framework that converts Isolation Forest decision paths into natural-language explanations so an LLM can refine anomaly detection through an uncertainty-aware dynamic ensemble. The approach adaptively weighs statistical and semantic signals based on instance-level confidence. Experiments on four real-world healthcare and finance datasets report an average 21% improvement in recall and a 10% gain in F1-score over baselines. The work also aims to improve transparency by grounding semantic narratives in statistical evidence.

arxivOct 29, 2026Score 76

anomaly detectionLLM reasoninginterpretability

Research

Formal Framework Targets Norm-Ability Alignment in Agentic AI

The paper proposes a formal framework for analyzing whether responsibilities in multi-agent systems align with agents' operational ability to fulfil them. It discusses design dimensions for describing normatively relevant processes, identifies issues such as unsatisfiable expectations and privilege escalation, and outlines repair strategies. It also examines how the relationship between operational process and institutional state affects the lifecycle of normative positions.

arxivOct 29, 2026Score 68

Agentic AIMulti-agent systemsAI governance

Research

CauTrace Diagnoses LLM Reasoning Failures via Internal States

A new white-box framework, CauTrace, analyzes why LLMs fail by tracking seven functional states for support and constraint processing, how representation changes propagate across Transformer blocks, and the timing and topology of commitment versus verification. Evaluations on Qwen and LLaMA-family models compare ordinary reasoning failures with paired adversarial perturbations. Failed traces show weaker Attention-MLP agreement on support-state updates and less verification relative to commitment, though exact signatures and timing shifts differ by architecture. Held-out repair and steering tests produce directional changes in process measures consistent with the framework for certain models and settings.

arxivOct 29, 2026Score 68

LLM failure analysismechanistic interpretabilityadversarial reasoning

Research

Agent-Conditioned Routing Elicits More Distinct Methods From One Frozen MoE Model

Agent-Conditioned Routing (ACR) runs parallel trajectories inside one frozen sparse mixture-of-experts model by applying persistent, untrained shared and private expert-eligibility masks while keeping the learned, input-dependent token router. On a screened 126-question mathematics and physics benchmark, ACR yielded about twice as many problems with at least two distinct primary methods among soft-correct solutions as the strongest temperature setting on GLM 5.3 (37.3 versus 18.7) and 1.67 times as many on GPT-5.6 Sol (45 versus 27), at matched output budgets. A blinded four-expert audit of 30 problems also found more multi-method solutions with ACR on both backbones, though correctness was lower. ACR adds about 160 KiB of mask state, and a three-problem microbenchmark stayed within 3% of batched temperature sampling on latency and peak VRAM.

arxivOct 29, 2026Score 74

Agent-Conditioned Routingmixture-of-expertsmulti-path reasoning

Research

FoG: Training KG Agents to Retrieve the Evidence Behind Their Answers

FoG is a framework for training knowledge-graph agents to retrieve the supporting triples behind factual answers, not merely produce correct answers. It combines evidence-grounded synthesis that hides explicit entities behind verifiable sub-questions, a KG interaction agent that records returned triples, and an evidence-coverage process reward that credits each supporting triple to the turn that first retrieves it. Experiments on three KGQA benchmarks show 3B–4B open-source agents outperforming substantially larger baselines in answer accuracy and evidence coverage, with zero-shot cross-KG transfer.

arxivOct 29, 2026Score 74

knowledge graphsLLM agentsevidence retrieval

Research Editor selected

Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation

This arXiv study uses causal interventions to test whether Gemma4-12B-it uses a shared internal numerical representation for text and image inputs and for addition and subtraction. It learns subspaces in the residual stream where swapping coordinates changes the model's answer, finding that reusing a text-derived subspace on images recovers only 53.2% of its causal effect, while a rotation and rescaling raises recovery to 91.4%. The analysis separates directions read by the output layer from hidden directions that still carry the number, and shows that swapping numbers in the hidden directions redirects the next digit in 77.6% of cases when the first digit is shared. The authors conclude that the model forms an abstract numerical concept upstream and unfolds it across time through relational geometry in different coordinate frames.

Why it matters: It offers mechanistic evidence about how a multimodal model represents numbers internally, distinguishing shared latent geometry from output-digit preparation.

arxivOct 29, 2026Score 84

mechanistic interpretabilitymultimodal modelsnumerical cognition

Research Editor selected

M4Bench Evaluates AI Coding Agents on Clinical EHR Derivations

Researchers introduced M4Bench, a benchmark of 28 clinical derivation tasks across 15 task families using MIMIC-IV and eICU, to test whether AI coding agents can reproduce EHR-derived variables and how clinician-reviewed agent skills affect results. In a Docker-isolated Codex campaign, skills raised reward by 0.136 over the no-skill baseline, with 26 of 28 tasks improving. Specification completeness accounted for most of the gain: an operational spec alone raised reward by 0.162 on an eight-task subset, while skills added only 0.024. The authors note the gains reflect partial agreement with reference derivations rather than completed tasks, and release tasks, skills, evaluation code, and sanitized per-run evidence as an open-source benchmark.

Why it matters: It provides a concrete benchmark and evidence on where coding agents help or fail in clinical EHR derivation, showing that precise specifications matter more than generic agent skills.

arxivOct 29, 2026Score 80

M4Benchclinical EHR derivationAI coding agents

Research

Concordat Enforces Typed Session Protocols on LLM Agents at Runtime

Concordat is an open-source Python library that treats an LLM agent as a finite-state protocol in the spirit of typestate and session types, and enforces that protocol at runtime. Each turn offers only tools legal in the current state, rejects out-of-protocol calls before side effects, allows human decisions only as requests, and charges token and tool-call budgets that split linearly across handoffs. Protocols are checked when defined, and a subclass is rejected if it could bypass its parent protocol. On tau-bench airline tasks with simulated customers pressuring agents to skip confirmation, a standard agent made changes without the required yes in 13 of 50 tasks with gpt-4.1 and 29 of 128 task-runs with Claude Haiku 5.5, while the Concordat agent made none, with no significant difference in task success; tau-bench’s reward missed most violations.

arxivOct 29, 2026Score 78

ConcordatLLM agentssession types

Research

Task Advancement Memory improves long-horizon GUI agents

A new arXiv paper introduces Task Advancement Memory (TAM), which treats verified task progress as both execution state and supervision for GUI agents. Task Advancement State tracks how verified progress changes remaining goals so agents can build on earlier results across applications, while post-task supervision abstracts interactions that produced verified local outcomes and keeps useful operations from both successful and failed tasks. On MobileWorld GUI-only tasks, Advancement State alone raises success from 12.0% to 28.2%. Full TAM reports a 17.6% relative gain in task success over the strongest evaluated baseline, a 38.7% reduction in lifecycle tokens per solved task, and state-of-the-art success on SPA-Bench single-app English Level 3.

arxivOct 29, 2026Score 76

GUI agentsTask Advancement Memoryagent memory

Research Editor selected

TeamGym benchmarks LLM agents on team decisions with distributed information

Researchers introduced TeamGym, an environment for evaluating and training language-model agents on team decision-making tasks built from legal, finance, medical, and software engineering corpora. TeamGym uses a method to identify "pivotal information"—a minimal set of facts that is individually necessary and collectively sufficient for a correct decision—so it can control how information is distributed and measure whether agents integrate it. Across 8 language models and 12 source corpora, distributing pivotal information lowered success by 22.9 percentage points versus giving it to a single model, and requiring agents to extract that information themselves lowered success by a further 19.9 points. Reinforcement learning on final-decision loss improved decision quality and produced emergent coordination in participation, speaker selection, and communication.

Why it matters: It provides a concrete benchmark and training method for a key multi-agent weakness: sharing and integrating distributed information.

arxivOct 29, 2026Score 82

TeamGymmulti-agent systemsLLM evaluation

Research Editor selected

Timestral open time series models claim top spots on forecasting benchmarks

Timestral is a family of four probabilistic time series foundation models with 5M, 28M, 123M, and 209M parameters. The design uses Decoupled Horizon Forecasting to separate prediction tokens from causally encoded context, a covariate framework for known-future covariates, and a two-stage pipeline that pretrains only on large-scale synthetic data before post-training on mixed real and synthetic data with longer context. The report says Timestral-209M and Timestral-123M rank first and second on every metric of GIFT-Eval, TIME, and fev-bench. Weights for all four models are on Hugging Face, with inference code on GitHub.

Why it matters: Open model weights plus a claimed new state of the art on major time series benchmarks give researchers and practitioners a concrete forecasting advance to evaluate.

arxivOct 29, 2026Score 80

Timestraltime series forecastingfoundation models

Research

CrossProfile shows rarest and relevance memory heuristics fail for LLM group decisions

The paper introduces CrossProfile, a diagnostic benchmark of 60 templates and 480 tasks with verified answer keys that crosses ownership of supporting, decoy, and irrelevant facts while holding their text fixed. On Qwen2.5-7B, Llama-3.1-8B, and two DeepSeek models, both rarest-first and relevance-based memory selection fail in predictable ways, and accuracy tracks the balance of evidence retained rather than the selection policy. A relevance filter loses 23 to 31 points to rarest-first selection in the classic hidden-profile setup because decoys stay on topic, while rarest-first is no better than random on average. The decision model also overweight facts attributed to many agents, and removing or randomly reassigning those tags raises accuracy by 13 to 24 points; full transcripts score worse than no notes.

arxivOct 29, 2026Score 78

CrossProfileLLM group decisionsmemory selection

Research

Temporal variability score shows high-variability knowledge improves faster late in pretraining

Researchers define a temporal variability score (TVS) as the coefficient of variation of an item’s negative log-likelihood across point-in-time language models trained on web snapshots from different years. Using six benchmarks and six models from multiple open-source families, they find that low- and high-TVS items are learned similarly early, while high-TVS items improve faster later in training. At the end of pretraining, high-TVS items outperform low-TVS items in 34 of 36 model-benchmark settings, with an average advantage of 7.3 chance-normalized accuracy points. The pattern is strongest on knowledge-intensive tasks, even for items of similar difficulty.

arxivOct 29, 2026Score 76

LLM pretrainingknowledge acquisitiontemporal variability score

Research

KPPax converts KPP chemistry mechanisms into differentiable JAX code

KPPax is a kinetic preprocessor that turns mechanisms written in the Kinetic PreProcessor (KPP) language into differentiable, just-in-time compiled, and vectorisable chemistry code in JAX. It evaluates rate laws with sparse operations in log space and integrates stiff systems using diffrax implicit solvers and forward- and reverse-mode adjoints, producing exact sensitivities without hand-derived adjoint code. On the POLLU benchmark, a Fisher information matrix from an automatically differentiated Jacobian predicted which of 25 rate coefficients a flow-tube-like design can identify, and noisy synthetic estimation matched the predicted uncertainties of the nine identifiable coefficients to within about 30%. A neural-network emission-inventory correction, trained offline on per-scenario 4D-Var retrievals and fine-tuned online through the chemical model, reduced biogenic correction error by a factor of 2.4–5 and recovered low-temperature emission information lost by the individual retrievals.

arxivOct 29, 2026Score 68

KPPaxdifferentiable ODEsatmospheric chemistry

Research Editor selected

FlashMap: Flow Map Drafter Achieves 6.14x Faster LLM Decoding

Researchers from Polymathic AI, Princeton, and partners released FlashMap, a speculative decoding drafter using continuous flow maps to draft 16-token blocks in one pass. On Qwen3-8B, it reports up to 6.14x lossless speedup on GSM8K, outperforming retrained DFlash on all tested benchmarks. The 0.5B-parameter drafter is available on Hugging Face with vLLM and SGLang plugins.

Why it matters: Offers a concrete, novel AI development with significant speedup claims and open-source availability, valuable for practitioners optimizing LLM inference.

MarkTechPostOct 28, 2026Score 82

speculative decodingLLM accelerationflow-based models

Research Editor selected

UC Berkeley's LEANN Vector Index Shrinks a 60M-Chunk RAG Index from 201 GB to 6 GB

UC Berkeley's Sky Computing Lab released LEANN, an open-source vector index that stores a pruned graph instead of full embeddings and recomputes embeddings only for nodes on the search path. Benchmarks show a 60 million chunk Wikipedia index dropping from 201 GB to 6 GB, and the paper won Best Research Paper at MLSys 2026. LEANN matches HNSW downstream accuracy on NQ at 25.5% while using 4 GB instead of 188 GB, but retrieval is slower: 2.48 seconds versus 0.05 seconds. The MIT-licensed project is designed to make semantic search and RAG feasible on laptops and personal data.

Why it matters: LEANN offers a practical storage-saving approach for RAG and vector search, with a clear tradeoff between dramatically reduced index size and slower query latency.

MarkTechPostOct 28, 2026Score 84

LEANNvector searchretrieval-augmented generation