All AI updates

Self-learning of large language models is incentivized through reinforcement learning

Through reinforcement learning (RL) and fine-grained rewards, researchers have developed an interface called BookNavigator to help large language models (LLMs) autonomously retrieve, learn, and apply knowledge from long reference materials to solve mathematical problems.

SteerDistill: Closed-loop teacher guidance for quantifying LLM agents

SteerDistill quantizes LLM agents through closed-loop teacher guidance, achieving an average score 1.67-2.18 points higher than the strongest quantization baseline on W4A16, a success rate of 1.84 percentage points on τ^2-Bench, and reducing training GPU hours by 67%.

TOME-Math: A mathematical proof supervisory corpus based on structured textbooks

TOME-Math transforms 2,118 university mathematics textbooks into approximately 380K structured proof supervision records, preserving chapter order, proposition-proof links, and dependency references, used for fine-tuning to improve the mathematical reasoning ability of language models by 16.7 percentage points, surpassing all three public instruction datasets. This corpus will be released to promote structured mathematical supervision research.

MIT researchers have developed new technology to accelerate and enhance the credibility of AI models for climate prediction.

MIT researchers have developed a training technique that significantly reduces data volume and improves the performance and adaptability of climate simulators (AI models), enabling them to quickly simulate the impact of human activities on future climate in seconds, such as predicting temperature or sea level changes, thereby more reliably testing thousands of climate scenarios and identifying unintended consequences of regional interventions. The algorithm generates optimized synthetic climate scenario data instead of relying on future emission scenarios based on real-world trajectories, significantly shortening production time and allowing the model to accurately predict scenarios not present in the training data.

One Table Benchmark launched: a benchmark test for general table models in generation, interpolation, and prediction tasks

The One Table Benchmark tested the table generation, imputation, and prediction capabilities of 46 models on 56 datasets, finding that these tasks share conditional information but have different fidelity and target interactions. Fixed models exhibit a trade-off between utility and fidelity, while frozen base models perform outstandingly in prediction and numerical imputation, indicating that unified table models have the potential to improve classification reconstruction and joint sampling.

Embodied First Exam Released: The First Robot Control Evaluation Framework for Coding Agents

Embodied First Exam has been released, including 35 sets, 47 robots, and 1,156 tasks, to evaluate the control capabilities of coding agents under physical world constraints, with GPT-6 Astra in Codex achieving a success rate of 34% in 456 core tasks under standard settings.

How does Amazon SageMaker HyperPod detect and recover silent data corruption?

Amazon SageMaker HyperPod detects silent data corruption (SDC) by adding three checks in the PyTorch DDP training loop, including non-finite gradient checks, loop replay audits, and weight fingerprint comparisons. When issues are detected, suspicious nodes are isolated and replaced by HyperPod, having been tested and verified on a 4 ml.p5.48xlarge instance cluster.

How Databricks Detects Credential-Driven Attacks

Databricks detects credential-driven attacks by analyzing audit logs and Genie, finding stolen credentials listed from new IPs in secret scopes, directories, and repositories within minutes, then reading secrets containing service principal credentials, and subsequently having the service principal create functions and join admin groups, these non-linear event sequences reveal intrusion behavior.

MACaQ: A Fast, Model-Agnostic Calibration Quantization Method

MACaQ is a post-training quantization method for MXFP4 W4A4, which reduces costs while maintaining accuracy in language and diffusion models by using online activation transformations and Hessian-aware refinement to independently quantize each linear layer. MACaQ achieves only a 2.3% end-to-end overhead increase on Qwen3-32B with vLLM, has a lower perplexity than DuQuant++ and MR-GPTQ, and reaches the best perplexity of Llama-3-8B in 11 times shorter quantization time.

Automatically update AWS Agent Registry using a push-based architecture

AWS CloudTrail captures the UpdateAgentRuntime API call event for AWS Agent Registry in Amazon Bedrock AgentCore, Amazon EventBridge routes the matched event to an AWS Lambda function, and the Lambda function queries the current tools from the MCP server and updates the matched Registry record.

Google Research has released a wearable device-driven CKM risk assessment framework

Google Research releases a wearable device-driven CKM risk assessment framework, predicting future health trajectories through an explainable health age model (CHA). The framework is trained on wearable data from 2.2 million weeks using the SensorWM model, providing personalized 12-week counterfactual health simulations combined with language models to generate clinically validated guidance plans. The CHA model integrates modifiable indicators such as heart rate, steps, and sleep to predict disease risk over 68,000 person-years, improving the predictive performance of clinical indicators.

FINCOSMOS: An open-world evaluation paradigm for intelligent agents in the financial domain.

FINCOSMOS has introduced an open-world evaluation paradigm for evaluating intelligent agents in the financial field. In 150 simulated user interactions, no session was accepted because all sessions failed during program checks, with the agent reporting only 54-64% of the undisclosed details. Human experts attributed 88.33% of the failures to the agent.

Using NVIDIA DSX AI Factory to Verify GPU Platforms through Runtime AI Checklist

Using NVIDIA DSX AI Factory, you can track actual AI workloads running on NVIDIA Dynamo, Kubeflow, or Slurm platforms, exceeding the visibility provided by platform configuration; application teams independently define and update workloads, and deploying a package cannot determine the workload components running on a node.

Google DeepMind AI models improve hurricane forecast accuracy, but human meteorologists remain irreplaceable

Encoder Separation of vLLM: Expanding Multimodal Service Capabilities through EPD

vLLM extends multi-modal service capabilities through the EPD (Encoder/Prefill/Decode) separation technology, supporting independently deployed visual encoders. vLLM v0.32.0 includes MiMo and XPU support, and the metadata-only request feature of DeepSeek V4.1 Flash has been open-sourced.

WikiTrace evidence chain memory system alleviates memory degradation in large language model agents

"WikiTrace Links Each Distillation Experience to Its Source Trajectory and Reasoning Chain, Allowing Large Language Model Agents to Verify and Correct Memory Before Action, Improving Performance and Reducing Degradation Caused by Memory in Five Intelligent Agent Benchmark Tests. The System Stores Each Experience as a Condensed Insight and Links It to a Summary of the Reasons for Its Success or Failure and the Original Trajectory, Enabling Agents to Supplement Missing Details or Correct Insights When Executing New Tasks Through Linking Tools. The Results Show It Outperforms Existing Memory Methods and Baselines Using the Same Insights"

InfraRSI: Utilizing recursive self-improvement training infrastructure

InfraRSI transforms open-ended human instructions into validated training optimization through a five-stage workflow, achieving up to 2.75x acceleration in four single-node and multi-node dense as well as Mixture-of-Experts (MoE) LLM workloads, and increasing the average model FLOPs utilization (MFU) from 17.4% to 30.0%, surpassing existing tuners and coding agents.

World In-Context Learning for Agentic Robot Manipulation

World In-Context Learning (WICL) method freezes the model, practices tasks in simulators, and writes feedback into the context rather than weights, enabling robots to learn operational skills from a single demonstration. RoboScript implements WICL, achieving an 83.4% success rate on 50 tasks in RoboTwin 2.0 with no gradient updates from a single demonstration, surpassing the strongest training strategy by over 20 points, and supports cross-visual distribution generalization, improvement through reflection, base model extension, and cross-platform memory migration.

The DexForesight framework enhances dexterous manipulation performance through implicit spatial foresight.

The DexForesight framework implicitly learns spatial foresight by injecting a dynamic perception world representation into the latent space of dexterous strategies, increasing the success rate of basic strategies from 49.4% to 60.5% on 11 DexJoCo simulation tasks and achieving a 58.3% success rate on 3 real robotic dexterous operation tasks.

MMDM: Large-scale Multi-task Decision Model Evaluation Benchmark

MMDM evaluated 28 decision models, showing that most models perform well on common knowledge but poorly on interdependent options, constrained decision-making, and multiple-choice tasks. For example, Perplexity Decider-V1.2 achieved 78.75% accuracy on single-choice knowledge, but only 53.95% and 47.95% respectively on option logic and constrained planning, while the best accuracy of all models on hard multiple-choice was only 36.81%.