Google Launches 'Sleep Framework' Continuous Learning Solution, Solving Large Model Knowledge Solidification and Forgetting Problems Through a 'Wake-Sleep' Cycle; The Model Processes Input When Awake, Integrates Memory and Self-Improves When Asleep, Experiments Show It Can Handle Millions of Token Contexts While Maintaining Stable Performance, AIME Benchmark Accuracy Improves
Hugging Face Daily Papers6565Research formalizes benchmark evaluation through difficulty measures to identify measurement mismatches and coverage gaps, proposing diagnostic metrics and rewrite algorithms to improve benchmark coverage; analyzing 25 popular benchmarks across 7 fields reveals that most implicitly overestimate specific difficulty areas or fail to cover most of the difficulty axis, limiting model improvement monitoring capabilities; introduces BenchFill algorithm to effectively reduce local coverage gaps.
When AI models process different languages, token consumption can vary up to 17 times, with English and other mainstream language users incurring lower costs. The implicit 'tokenizer tax' results in users of less common languages paying higher fees, affecting response speed and context capacity. AI labs need to publish evaluation results for each language and improve tokenizer design to narrow the gap.
25 GPT-6.1 Terra agents run in Minecraft for 3 hours, aiming to build a prosperous civilization. They form a pseudo-government, establish rules, and develop complex coordination patterns. The agents initially focus on cutting down trees, then form a hierarchy, and communicate through approximately 5,900 messages.
Hugging Face Daily Papers5252WeVisCut coordinates video editing tools, recovers from failures, and reduces token usage through a compact strategy trained on 9B parameters, while maintaining output quality. The system employs a four-stage data pipeline, combining editing inversion and validation replay, to generate over 130K training examples from 3,750 finished videos for training a single strategy.
Hugging Face Daily Papers3838Soft-MTP removes noise gradients residuals in multi-label prediction (MTP) using KL recovery techniques, improving perplexity, speculative decoding, and reasoning ability without increasing cost.
Epoch AIFeaturedSame story · 2 outlets8888Epoch AI's experiment shows that the open-source weight model GLM 5.3 released by Z.ai in August can autonomously construct web attacks targeting six Firefox versions released between September 17 and October 26, exploiting publicly available but not yet pushed security fixes, where visiting the web page causes local files to be read and transmitted.
How the accounts differ:The Decoder mentions the cost of 1000 dollars and that the vulnerability was fixed in Firefox 159.0, while Epoch AI does not mention these details.
Why it matters: Epoch AI replicates patch window attacks using open-source weight models, providing a cost level of about 1000 US dollars per version, available for evaluating the security boundaries of open models.
Epoch AI6565Epoch AI analyzes global AI chip sales data, measuring computing capacity in H100 equivalent units, showing exponential growth each quarter, indicating computing capacity doubles every 9 months.
Databricks Blog2828When evaluating Postgres management services, focus on infrastructure operations and maintenance, disaster recovery, and compatibility, rather than just the feature list. Key validation points include: operational scope (patches, upgrades, tuning permissions), high availability (automatic failover, cross-availability zone switching), backup and recovery (PITR, retention period), compatibility (supported PostgreSQL versions and extensions), migration paths and costs. Red flags include managing only virtual machines, no clear RPO, critical extensions not supported, or data export restrictions.
Amazon ScienceAmazon Science combines knowledge distillation (KD) and the Medusa framework to optimize the travel platform LLM. After fine-tuning, KD training, and Medusa head modification, the TinyLlama 1.1B model achieves performance close to the 70B parameter Llama-3.1 teacher in AI travel planning, intelligent filtering, and accommodation description generation scenarios, with costs reduced by two-thirds and speed increased by 10-20 times.
AWS Machine Learning Blog4242MoEngage transforms 400 internal APIs into permission-controlled intelligent agent tools through Amazon Bedrock AgentCore Gateway, solving multi-team API integration challenges. This solution eliminates the need to set up MCP servers or configure networks, reducing the time to launch new APIs from days to hours while meeting constraints such as keeping APIs within the VPC, controlling identities from the server side, and ensuring tool definitions do not occupy the reasoning context window.
Meta EngineeringFeatured6666Meta announced that Catching JiTTests has been put into internal production, using LLM to automatically generate executable tests for diffs to be merged, exposing behavioral differences before the code is landed
Why it matters: Meta has disclosed the production deployment details and quantitative results of JiTTests, available for evaluating the implementation of LLM-generated tests in ultra-large-scale codebases.
NVIDIA Technical Blog3535NVIDIA significantly accelerates machine learning atomic potential energy calculations by optimizing Tensor Core performance. Its technology can improve the computational efficiency of quantum mechanics methods such as Density Functional Theory (DFT), balancing cost and accuracy. This solution is suitable for large systems or long-time simulations, but the specific parameter scale, benchmark scores, and availability (open source/closed source/API) were not provided in the original text.
Hugging Face BlogProject ProtoStar is announced, aiming to unlock the potential of publicly funded scientific research by making datasets, models, and code more accessible and reusable. The project brings together universities, research institutions, industry, and the open science community, supporting the release of scientific achievements in AI-compatible, machine-actionable formats, and demonstrating its value through a series of public challenges. The first challenge, hosted by Carnegie Mellon University, will launch on November 2.
Lean founder Leo de Moura discussed how AI can make formal verification more accessible and potentially change software engineering. AI-generated code needs to be verified for correctness using formal verification, and Lean can be used to write machine-verifiable proofs, with two verifiers already available. AI can assist in implementing code and automatically proving its properties, such as when Kim Morrison ported zlib to Lean and proved it to be lossless compression.
Today is auspicious for rest at ISHO Today is auspicious for rest AI Sleep Technology Ecosystem Product Launch Conference announced its strategy, focusing on precise monitoring, scientific intervention, and energy recovery, planning to collect over 1000 cases of gold standard clinical PSG data within a year, and extending the research scope from nighttime sleep quality to daytime energy status.
DAYJOB: Tax evaluates the performance of AI agents in 50 professional tasks across 8 tax areas, requiring agents to complete tax filing, compliance, and advisory work. Claude Sonnet 5.5 leads the rankings with a pass rate of 10.8%, while Claude Opus 5.5 has a pass rate of 9.2%. The benchmark includes 1,789 workspace files, covering various tax scenarios and complex materials, created and reviewed by tax experts.
Physical Superintelligence (PSI) claims to have achieved results on 5 open problems in quantum information theory with the help of its self-developed AI research system Emmy. Papers, Lean formal proofs, and verification code have been released on GitHub for review.
Why it matters: Readers can see a new company challenging quantum open problems with a process that separates generation from formal verification, and learn about its team and funding background.
Hugging Face Daily PapersThe Prox model accelerates expert mixture inference through top-k channel selection and weighted reader, while maintaining accuracy without calibration data. Under 13% and 18% budgets, the top-k version of Prox outperformed or matched the threshold version in 40 out of 43 comparisons, including all 23 generation tasks, and avoided the repetitive loop crashes caused by calibration thresholds on Gemma 4.
Momenta CEO Cao Xudong first publicly defined the large model definition chip, which improves performance 10 times compared to general chips and can support billion-parameter models to achieve large-scale L4, driving breakthroughs in physical AI applications such as robots. XHEART New Chip Journey released a new X7 Ultra large model chip, specifically designed for large models, featuring a super-large core NPU architecture and Armv9 automotive-grade CPU.
Hugging Face Daily Papers6565The Sensorimotor Alignment (S²) metric finds that the alignment degree between vision and motor representations of embodied agents is positively correlated with task success rates, verifying the applicability of the Platonic representation hypothesis in the robotics field. In simulation and real robot tests, models with high S² (such as Vision-Language-Action and World Action Models) perform better and can explain the phenomenon where the VLA model ignores instructions and relies on visual features, serving as a low-cost offline proxy metric.
MolCell AI's MMDesign platform completes 18 design projects with a success rate of 94.7%, hitting all six GPCR targets among them, with the best affinity reaching nanomolar level; the platform uses a 'generate-screen-verify' workflow, generating tens of thousands of candidate molecules in computational space and testing only 9-50 to achieve success, with an overall candidate hit rate of 30.7%, demonstrating efficient and precise molecular design capabilities.
Zhuji Dynamics releases the FluxVLA Engine V2 technical report, achieving full-chain reuse of model training, evaluation, and deployment through unified configuration, modular decoupling, and system optimization. It supports 14 mainstream strategies, with inference path frequency improved by 2.31–9.64 times, validated in closed-loop execution on 6 real machines including ALOHA and dual-arm Franka. Online reinforcement learning boosts SmolVLA success rates by 10.0–15.4 percentage points.
Zhuojì Power updates the second technical report of the open-source engineering base FluxVLA Engine. V2 supports 14 strategies for LIBERO evaluation on the simulation side, increases model path frequency by 2.31–9.64 times on the inference end, and verifies deployment links on real hardware platforms such as ALOHA and dual-arm Franka. Online reinforcement learning boosts SmolVLA's success rates in LIBERO-10 and RoboTwin Clean tasks by 10.0% and 15.4%, respectively.
Hugging Face Daily PapersEmbeddingGemma 2 is an open-source multimodal embedding model that unifies text, image, audio, and video representations in an efficient and resource-flexible design, and performs excellently in benchmark tests. It has established a new Pareto frontier in embedding benchmark tests for all modalities, outperforming professional models and other models that are two to twelve times larger in scale. The model supports selective modality encoder loading, adjustable visual token budgets, truncatable embedding dimensions, and quantized weights to adapt to resource-constrained environments.
Hugging Face Daily PapersFalcon is a causal audio-visual target speaker extraction model that extracts the voice of the target speaker from the mouth video of the target speaker in noisy overlapping speech, combined with a single multi-scale time-frequency separator and an encoder spatial speaker contrast loss, reducing interference speaker leakage without increasing inference cost.
Hugging Face Daily Papers6565DTCA performs label-level credit assignment through confidence enhancement and entropy calibration, improving the mathematical reasoning ability of diffusion language models. The framework, when tested on LLaDA-8B-Instruct, shows significant performance improvement in mathematical reasoning under different generation budgets, with the MATH500 metric increasing by 9.0 points and the GSM8K metric improving by 5.3 points.
Hugging Face Daily PapersRSI Arena demonstrates that autonomous AI intelligence can independently use human feedback to train and iteratively improve language models. Most agents achieved preference gains in the second phase, although benchmark test improvements were limited. Ten cutting-edge models were used as training agents, starting from NVIDIA Nemotron 3.5 Lightning 30B-A3B Base, which independently selected data, implemented training, and selected checkpoints in the first phase within six days, using a 64-GPU cluster and 300 USD API credits.
The first batch of extreme weather AI models for the Fengyun meteorological satellites were released, with the typhoon intelligent cloud map interpretation model interpreting every 15 minutes and generating typhoon analysis text in 27 seconds; the short-term warning model for strong convection, combined with observations from Fengyun-4 satellite, will advance the warning time to four hours.
Replit BlogViBench 2.0 Released, used to evaluate the ability of models in long-term software construction, including 17 products, 241 feature requests, and 544 test plans, top model GPT-6.1 Sol.
AWS Machine Learning Blog5858Amazon Bedrock AgentCore provides modular services such as runtime, gateway, identity, memory, policy, registration, and evaluation, supporting large-scale agent construction, deployment, and operation. The enterprise-grade agent platform blueprint based on AgentCore adds enterprise-level constraints such as ownership, account boundaries, software delivery, observability, cost governance, and retirement, achieving controlled access through inference gateway and tool gateway, and including an evaluation process to ensure agents are validated before production deployment.
LangChain BlogOpenWiki makes agents about 5 times more accurate on cross-source connection issues than single search, adding evidence images increases accuracy by another 27% and doubles citation consistency.
Hugging Face Daily PapersThe EvidenceConnect framework generates comparative, aggregated, or chained visual evidence queries by synthesizing image region thumbnails, enhancing the evidence connection capability of VLMs. The ZwZ2 model is trained based on this framework, without targeting any downstream benchmarks but improving fine-grained and spatial perception, and achieving progress in embodied spatial reasoning, multimodal long-context understanding, multimodal agent tasks, and general visual reasoning
Hugging Face Daily Papers6565Multiscale Policy Optimization (MPO) improves the training of long-horizon LLM agents by aligning credit assignment and policy updates at the episode scale. Experiments on search, tool use, and automated tasks using 4B and 30B-A3B models show that MPO has sustained improvements over strong RL baselines.
Simon Willison6262Claude Opus 5.5 significantly improves 3D modeling, music generation, and game design, making it easy for AI to build fun games. Developers created four HTML/JavaScript games, including Tree Appreciator and Emberwick, with a single prompt. OpenAI expects GPT-6 Astra to rival Opus 5.5's fun by the end of the year.
MIT Technology Review · AI00MIT Technology Review analyzes the latest advancements in industrial AI autonomy and human control, emphasizing the importance of human oversight during automation processes.
