iCreatAI News
Back to main site
Skip to content
Only First party#Data / training
Nov 9Mon
  1. Databricks Blog36

    Accelerate GPU training on Unity Catalog tables with UCDataset

    Databricks released UCDataset in Beta on AI Runtime, a native PyTorch data loader that streams Unity Catalog tables directly to every GPU in a training job without exports, copies, or custom loader code. On a simulated production workload streaming a 300 GB feature table across 8 NVIDIA H100 GPUs to train a recommendation model, UCDataset cut training time by 10x on average compared with loading data directly from Parquet. The Beta reads Delta tables in Unity Catalog, applies filters and deletion vectors as data is read, and enforces Unity Catalog permissions, though it cannot read tables with row filters or column masks.

  2. Hugging Face Blog36

    Nemotron-Personas-Finance-Japan: A Synthetic Dataset Capturing Japan's Demographics and Financial Context

    Elith and NVIDIA released Nemotron-Personas-Finance-Japan, an open synthetic persona dataset built on official Japanese statistics and household finance and financial literacy data, generated with NeMo SDG-PGMs and NeMo Data Designer. It contains 9M personas (1M records x 9 personas) in natural Japanese across 50 fields, released under CC BY 4.0 for commercial and non-commercial use. The dataset is designed for financial AI development such as conversational agents, customer simulation, model evaluation, and bias testing without using real customer data.

  3. Cloudflare Blog30

    Launching the Vite Ecosystem Fund

    Cloudflare is distributing the first $301,300 of the Vite Ecosystem Fund to 44 projects and maintainers, part of a $1 million commitment made when VoidZero joined Cloudflare. The post argues that AI has made open source funding urgent by increasing AI-authored pull requests and maintainer burnout, and it describes the fund's selection panel and stipend structure of up to $4,000 per month at $50 per hour for eligible core maintainers.

  4. Lovable Blog38

    How we moved our frontend off Firestore and built Lovable Realtime

    Lovable built its own WebSocket-based realtime service, Lovable Realtime, to replace browser-side Firestore listeners, letting its Go backend control payload shape and re-evaluate authorization continuously across Firestore and Spanner data. The service handles peaks of over 2 million concurrent subscriptions and delivered about 2.61 billion data and patch frames in a month, with update latency of 66 ms at p50 and 249 ms at p99. The rollout used shadow traffic and a gradual feature-flag ramp to 100% of users without a user-facing incident.

  5. NVIDIA Newsroom25

    NVIDIA Announces New Planned Investments in Korean Startups at the MSIT-NVIDIA AI Startup Partnership Forum

    Korea's Ministry of Science and ICT and NVIDIA hosted the MSIT-NVIDIA AI Startup Partnership Forum at COEX, where Vice Minister Ryu Je-myung and NVIDIA leaders met with over 20 Korean startups spanning foundation models, physical AI, AI infrastructure and AI semiconductors. NVIDIA announced an investment in Point2 Technology, an AI interconnect company building RF-based interconnects for AI cluster scale-up, with additional investments planned in Motif Technologies, RLWRLD and Upstage. The planned investments build on NVIDIA's work with Korea's startup community, including more than 1,000 Korea-based members of NVIDIA Inception.

  6. MIT News AI58

    A single AI model simulates the behavior of many substances, from sliding sand to folding proteins

    Researchers at MIT, UBC and elsewhere have developed WorldParticle, a deep-learning system that represents materials as Lagrangian particles and predicts their motion one step at a time, learning underlying dynamics rather than relying on hand-coded solvers. The model was demonstrated on six physical phenomena, including deformable objects, granular materials and proteins, with higher accuracy than numerical methods and modeled protein structural changes about 100 times faster than numerical simulators. The work will be presented at ACM SIGGRAPH Asia.

  7. Artificial Analysis Articles42

    Introducing the Artificial Analysis Decision Index v2

    Artificial Analysis released Decision Index v2, with 45% of the Index built on private datasets (up from 25% in v1) and a new Decision Score that rewards both correctness and calibrated confidence. Drex Max v1.1 (Nace AI) and GLiDE v1.2 (Fastino Labs) both score 73, with Mammoth 27B (Baseten) at 72, while cost per 1M Tasks ranges from $14 for Celeris-1.1 Decision to $859 for Clef. The update adds AA-Paperclip (1,500 private documents), AA-Symmetry, revised AA-Omniscience-Decision and HLE-MultiChoice benchmarks, and EvalSafe-O*NET-AA, while reducing When2Call to 10% and removing WildGuardTest.

  8. Artificial Analysis Articles40

    Endpoint Accuracy Index v1.1: Tighter long context intervals and expanded model coverage

    Artificial Analysis released Endpoint Accuracy Index v1.1, which updates AA-LCR-25 to dataset version 1.1 with corrected answer keys, an updated equality-checker LLM and 25 repeats per question, narrowing the average 95% confidence interval from ±4.7 to ±2.6 points. Coverage now spans 103 serverless endpoints across six open weights models, adding GLM-5.3 Flash and Muse Spark Open and replacing DeepSeek V4 Pro with DeepSeek V4.1 Flash 1017, while GLM-5.2 leaves. Of the 103 endpoints, 24 score statistically below their self-hosted reference, with causes grouped as output token caps, reduced reasoning, tool call handling differences, and request limits or stream behavior.

Nov 7Sat
  1. MIT News AI38

    How to predict the behavior of molecules from scant data

    MIT researchers developed CheMeleon, a pre-trained foundation model built on the open-source ChemProp package that learns from descriptors of a million molecules in PubChem and can be fine-tuned to predict molecular properties from limited data. In tests across 58 cases it outperformed other foundation models and classical expert-designed models, and nine of the top 20 entries in an OpenADMET contest used it. The findings were published in the Journal of Chemical Information and Modeling by Zalte, Burns, Abreu, Green, and three BASF researchers.

  2. NVIDIA Technical Blog22

    Accelerating Reinforcement Learning Weight Refit with the Reshard Collective

    This post discusses how reinforcement learning post-training generates trajectories by running inference with the current policy model, and how disaggregated setups run inference and training in parallel on separate GPU resources for efficiency. Because the policy model is updated only on the training GPUs, the system must periodically refresh the weights on the inference side, and the post introduces the Reshard Collective to accelerate this weight refit.

  3. AWS Machine Learning Blog58

    SageMaker Inference now supports Multi-Instance GPU to maximize GPU utilization for model endpoints

    Amazon SageMaker Inference now supports NVIDIA Multi-Instance GPU (MIG) partitioning on inference component endpoints, letting you divide a physical GPU into up to seven hardware-isolated partitions and place a different model in each. Partitioning is declared via AcceleratorPartitionLayouts on the endpoint config and AcceleratorPartitionType on each inference component, with no new API operations or IAM actions. In testing, a mig-3g.40gb partition of an H100 delivered 53–89% more output tokens per second and 18–34% lower cost per million output tokens than a full A100-40GB above a workload-dependent concurrency crossover.

  4. Databricks Blog38

    Real-time fraud detection with Feature Views

    Databricks describes how Feature Views in its Feature Store let teams define a feature once and use the same definition for point-in-time-correct training data and managed online-serving pipelines, reducing training-serving skew and data leakage. The walkthrough follows a $900 payment from a Kafka stream through rolling and Sawtooth window aggregations, an on-demand deviation feature, model training and Model Serving, with Unity Catalog providing governance and lineage.

  5. Databricks Blog30

    How public sector organizations can govern AI at scale

    Databricks outlines why public sector AI projects stall on governance questions—model access, data residency, and spend visibility—and describes how its Unity Gateway provides centralized controls over which models, coding agents, and tools teams can use, with per-user or per-team spending caps, platform-level compliance policies, and geographic data residency for hosted foundation models. The piece argues that agencies which solve governance centrally before scaling adoption can match each mission to an appropriate model while keeping data, users, and budgets under control.

  6. Baseten Blog70

    Introducing Serverless Loops: frontier RL training for everyone

    Baseten announced the general availability of Serverless Loops, a Tinker-compatible training SDK for supervised fine-tuning and reinforcement learning on frontier open models at 256k context length, with one-click deployment to Baseten Dedicated Inference or weight download in Hugging Face Safetensors format. The release cites early users including NVIDIA, Datadog, LangChain Labs and Ambience Healthcare, and reports NVIDIA's post-trained Nemotron Super 3.5 VL improving intent classification accuracy from 76.8% to 94% and reaching 87% validation F1 on receipt extraction.

    Why it matters: Useful for readers weighing managed post-training options, as it pairs a GA launch with named early adopters and concrete reported results.

Nov 6Fri
  1. Hugging Face Blog58

    ToyBlocks: A controlled world for robot learning and evaluation

    ToyBlocks is an open simulation framework built around a robot arm and blocks on a tabletop that automatically generates tasks and demonstrations, using reinforcement learning and motion planning to produce skills such as pushing, flipping, and stacking. It supports studies of low-level action model generalization, high-level planning with vision-language models, agentic systems using tool calls, action-conditioned world models, and zero-shot sim-to-real transfer, with an initial release of 85,336 long-horizon demonstrations covering nearly 1,600 hours and 144 million frames. The authors present ToyBlocks as the first of their FoundationWorlds and plan a second release with a larger dataset, more skills, and integrated RL training.

  2. Artificial Analysis Articles40

    VIALS-AA: different paths to scientific image reasoning

    VIALS-AA analyzes how 19 models reason over scientific images, classifying 9,177 tool calls from its original test set: 38.3% were image viewing and 47.3% prepared or measured images. Image measurement made up just 5–6% of tool calls for GPT-6 Astra (max) and GPT-6.1 Sol (max), versus 25–27% for Claude Opus 5.5 (max) and Claude Sonnet 5.5 (max). The report also notes that Gemini 3.8 Flash (high) accounts for 86.7% of commentary/no-op calls, that more intelligent models tend to call fewer tools, and that cost per task varies by nearly 100× across models.

  3. vLLM Blog42

    VeRL-Omni v0.3.0: Async Training and Broader Multimodal Post-Training

    VeRL-Omni v0.3.0 adds asynchronous and decoupled RL execution paths across multimodal workloads, makes the TransferQueue/ReplayBuffer-based V1 trainer the default, and moves the rollout stack to vLLM 0.30.0 and vLLM-Omni 0.30.0. It includes end-to-end recipes for MiniMax H3 audio-video post-training with DiffusionNFT and MiniCPM-o 4.5 AVQA GSPO, where a 4×H800 experiment found separate-async about 9% slower per prompt at natural response lengths but roughly 6.7% faster at 1024 tokens. The release also expands model and algorithm coverage, adds Ascend and AMD ROCm Dockerfiles, and notes upgrade changes including V1 as default, named diffusion media artifacts, and fail-closed attention backends.

  4. AWS Machine Learning Blog29

    How Amazon Marketplace evaluates content quality with Amazon SageMaker AI

    Amazon describes a system that renders a full A+ Content page as a single snapshot and scores it with a fixed Vision-Language Model across four rubric dimensions, then fuses those scores with four OCR readability signals in a CatBoost classifier trained on expert annotations. On a 20 percent holdout set of 5,478 product detail pages from 10 categories, the fused model reached 0.94 accuracy and 0.95 F1, versus 0.80 and 0.84 for an averaged-score baseline, and SHAP values plus an LLM turn the results into prioritized recommendations. The pipeline is implemented on Amazon SageMaker AI using Processing Jobs, Inference endpoints, Feature Store, Training Jobs, Model Registry and monitoring, with a diffusion model generating improved image assets.

  5. METR Blog78

    Independent Assessment of AI R&D Acceleration at Anthropic in October 2026

    METR conducted a high-access evaluation at Anthropic from August 12 to October 29, 2026, concluding that Anthropic is likely improving model capabilities on well-specified tasks at around twice its March 2024–October 2025 rate, with AECI benchmark scores growing at an estimated 1.9x the pre-October 2025 rate (95% CI: 1.5–2.8x). METR estimates that a majority (perhaps ~80%) of the current acceleration is due to AI, found no clear bottlenecks to continued progress over the next 12 months, and cautioned that Anthropic redacted most of the evidence it found most informative.

    Why it matters: It offers a rare public accounting of how frontier AI R&D acceleration is measured, including explicit uncertainty ranges and disclosure of what redactions prevented the public from judging.

  6. Apple Machine Learning Research38

    Faster Retrieval for Better, More Efficient RAG

    Apple researchers present amortized maximum inner product search (MIPS), training small neural networks offline on a fixed key database to learn the support function of the key set, which is convex and whose gradient returns the best-matching key. SupportNet learns per-cluster support functions to route queries more accurately than centroid-based routing, while KeyNet rewrites queries before a standard IVF index; on Natural Questions, SupportNet reached 95% routing recall about 1.9× faster than centroid routing, and KeyNet cut the time to 99% recall by up to ~8×. The work was presented at ICML 2026.

  7. Anthropic Research58

    AI for scientific discovery: How researchers use Claude

    Anthropic researchers used privacy-preserving classifiers to label 207,049 Claude and Claude Code research sessions (August 2–September 30, 2026) by field and task, finding heavier use in formal sciences like computer science, mathematics, and statistics than in health and lab-based physical sciences, with data analysis and writing dominating. Claude use rises with a constructed "theoretical AI capability" score, which correlates with reinforcement-learning suitability and cognitive work and negatively with physical, interpersonal, or regulated work. Scenario modeling suggests current capabilities could double research output in some fields, and full automation of non-physical work could raise output up to 15-fold in fields like computer science and statistics, versus three- or fourfold in bench sciences.

  8. Databricks Blog58

    From Delta Sharing to OpenSharing: what changed, what's next

    In June 2026 Databricks and the Linux Foundation announced that Delta Sharing has evolved into OpenSharing, an open, vendor-neutral protocol under the Linux Foundation AI & Data Foundation that extends sharing beyond tables to volumes, agent skills, and AI models, and adds Apache Iceberg REST Catalog support and on-premises or private cloud reach. Existing Delta Sharing clients and integrations remain compatible, and enabling OpenSharing on a Unity Catalog metastore is a configuration step rather than a migration. Databricks also introduced capabilities on top of the protocol, including SecureConnect (generally available), Global Distribution (private preview), Genie Agent Sharing (beta), and Lakebase table and change data feed sharing (public preview).

  9. Databricks Blog22

    Which Data Collaboration Tool: OpenSharing, Clean Rooms, Marketplace

    This Databricks blog compares three data-sharing methods: OpenSharing, which gives recipients direct read-only access to live assets such as tables, views, notebooks and AI models; Clean Rooms, where up to ten parties combine data and run only mutually approved code without exposing full underlying datasets; and Marketplace, a public storefront for discovering and publishing data and AI products. It advises choosing based on the relationship with the recipient and notes the tools are not mutually exclusive.

  10. Waymo Blog36

    Read more: How the Waymo Driver Manages Severe Winter Weather

    Waymo describes how its Driver handles winter conditions, using cameras, radar and lidar with heaters and micro-wipers to keep sensors clear, and measuring real-time traction to adjust steering, braking and acceleration. The company says it may slow, limit or pause service when weather exceeds the boundaries defined by its Safety Framework, drawing on fleet-shared road data, forecasts and local officials. Waymo says it has accumulated tens of thousands of real-world miles in snowy regions including Upstate New York, Michigan's Upper Peninsula and the Sierra Nevada, and will conduct closed-course evaluations and heavy snowfall data collection in Alaska this season.

  11. NVIDIA Technical Blog22

    Five Ways NVIDIA Co-Design Maximizes the Value of Every Byte in the AI Factory

    NVIDIA outlines five co-design approaches for maximizing the value of every byte in an AI factory, where high bandwidth memory (HBM) is described as the fastest, most expensive and most contested memory tier. The piece argues that bytes spent on redundant weight copies, idle sessions' KV cache or recomputed prefixes are bytes not producing tokens, shifting the question from how much memory a GPU has to how much useful work each byte delivers.

Nov 5Thu
  1. Amazon Science18

    Hamed Karimi wins ECML PKDD Test-of-Time Award

    Amazon senior applied scientist Hamed Karimi and coauthors Julie Nutini and Mark Schmidt won the 2026 ECML PKDD Test-of-Time Award for their 2016 paper on linear convergence of gradient and proximal-gradient methods under the Polyak–Łojasiewicz condition. The paper showed that the largely overlooked 1963 inequality can prove fast linear convergence for standard training methods even when many solutions work equally well, mapped how previously scattered weaker-condition results relate, and extended the analysis to nonsmooth landscapes and algorithms including coordinate, sign gradient, and stochastic gradient descent. It has been cited over 2,000 times and is used in graduate textbooks.

  2. Baseten Blog22

    Baseten joins the NVIDIA Nemotron Coalition to advance custom AI in production

    Baseten has joined the NVIDIA Nemotron Coalition, expanding its collaboration with NVIDIA on open frontier models, post-training, RL environments, and dedicated inference. The post cites CodeRabbit's Nemotron 3.5 Lightning customization, which raised routing-policy agreement from 75.8% to 80.7% on 1,000 tasks and cut inference cost 50.4% at saturated throughput, and notes Harvey post-trains Nemotron on Baseten for legal workflows. Baseten Loops now supports Nemotron 3.5 Super VL and Nemotron 3 Ultra, with customers retaining ownership of trained weights.

  3. Cerebras Blog50

    What It Took to Build Wafer-Scale, and Why It Lasts

    Cerebras CEO Andrew Feldman rebuts a Culper Research short-seller report claiming its wafers wear out after about two years, saying the company does not use the PariPoser material and that its patented assembly keeps temperatures steady and even while flexible connections absorb thermal expansion. He cites TSMC qualification for 10 years of continuous operation at 105°C and an average junction temperature below 65°C across deployed CS-3 systems, arguing Arrhenius modeling with those inputs implies 33 years at maximum temperature and over a century at average. Feldman also says fewer than 0.5% of deployed CS-3 wafers have been decommissioned since 2024 and defends the company's 3-5 year depreciation schedule as conservative.

  4. Databricks Blog36

    Push-based CDC into Unity Catalog with Zerobus Ingest and Debezium

    Debezium Server 3.7 ships an open-source Zerobus Ingest sink, built by Databricks with NVIDIA's contribution of envelope mode, that pushes change events directly into Unity Catalog Delta tables without a Kafka cluster between capture and the lakehouse. Envelope mode lands each full Debezium change event in a fixed nine-column table with source position and an idempotency key, and the sink waits for Zerobus to confirm durability before Debezium advances its offset, giving at-least-once delivery. Landed events can then be shaped into Type-1 current-state tables via AUTO CDC sequenced by source LSN, or Type-2 history tables.

  5. Anthropic Research66

    Investigating vault particles with Claude

    Anthropic's biology research lab reports the first 3D structure of a vault particle from a prokaryotic cell—a bacteriophage-derived vault built from 96 copies of a single protein, L05, solved by cryo-EM. Under researchers' direction, Claude searched protein databases, designed DNA constructs, optimized a size-exclusion chromatography isolation method that cut purification from about three days to roughly 90 minutes, and processed microscopy data, yielding a first structural model in 34 hours. The team also engineered altered and enlarged vault variants, and notes the work demonstrates using Claude across every step of fundamental biology research.

    Why it matters: Describes a concrete AI-assisted structural biology result, including the methods, speedups, and limitations the team encountered.

  6. Databricks Blog42

    The Enterprise Case for Open-Weight Models on Databricks

    Databricks field engineers describe seven pressures pushing enterprises away from single frontier-model providers, including insufficient returns on AI spend, frequent ranking changes, guardrail refusals, shifting data-retention policies, governance demands, harness effects on performance, and open-weight models now benchmarking comparably to frontier models at lower cost. They present Unity Gateway as a unified API for routing and governing open-weight, frontier, and external models, citing internal data showing average unique models per customer rose from 4.2 to 7.1 since January, with roughly 40% open-weight. The post also notes GLM 5.2 reaching about 40% of open-weight tokens on Unity Gateway a month after launch and Databricks' $3 billion July raise, partly for GPU capacity to host open models.

  7. Hugging Face Blog60

    Dharma Field Extractor: Any Document In, Structured Fields in JSON Out

    Dharma-AI introduces the Dharma Field Extractor, a 2-billion-parameter model built on DharmaOCR-LITE v1.1 with a LoRA adapter trained on 43,384 examples, which takes a document image plus either an open request or an exact schema and returns typed JSON, including null for fields not present. On a 546-document, 1,638-case benchmark it scored 0.834 field-level F1, above every frontier configuration tested (Claude Opus 5 at 0.816 to GPT-5.6 Luna at 0.741), while running self-hosted on a single L40S at a cost the company reports as 6 to 147 times lower than the APIs compared. The company frames the result as evidence that a small model trained close to a bounded task can beat a general frontier model on that task.

    Why it matters: Useful for teams weighing self-hosted specialized extraction against frontier APIs, though the benchmark and cost figures come from the vendor itself.

  8. Google Research52

    Agile Deliberation: How AI helps humans define the gray areas

    Google Research introduces Agile Deliberation, a framework that treats building subjective AI classifiers as an iterative conversation, splitting the task into concept scoping (breaking a broad idea into sub-concepts and exclusions) and concept iteration (using vision-language models to surface borderline images that challenge the user's definitions). In 18 live 1.5-hour user sessions across the subjective domains of "Healthy Food" and "Pay-to-Play" content moderation, the approach produced more consistent, higher-performing classifiers than zero-shot, automated collaborator, and manual deliberation baselines while reducing user fatigue and anxiety. The code and algorithms are open-sourced on GitHub, with a Colab/Jupyter demo available.

  9. AWS Machine Learning Blog38

    How Indegene is modernizing and scaling enterprise data quality in life sciences

    Indegene built an LLM-powered Data Quality Agent within its NEXT Unified Data Platform on AWS, using Amazon Bedrock, Strands Agents, and Amazon Bedrock AgentCore to generate context-aware data quality rules, validate them, and produce SQL and Python code under human-in-the-loop governance. The agent draws on internal enterprise and external reference knowledge bases via RAG to recommend technical and life sciences-specific business rules. Indegene reports the agent reduced rule definition and implementation effort from about 30 hours to about 4 hours per dataset and cut new table onboarding from 2–5 working days to 2–4 hours.

  10. Allen Institute for AI Blog38

    Ai2 and Google collaborate on a new generation of climate models

    Ai2 is launching a climate modeling initiative with multimillion-dollar, three-year funding from Google.org and technical support from Google Research to develop NeuralGCM+, an open-source model extending Google Research's NeuralGCM to simulate atmosphere and ocean together for faster, more reliable decade-scale climate projections. Ai2 will focus on climate trends over years to decades, such as how rainfall and water availability could change through mid-century in regions like the southwestern United States, while Google Research continues developing NeuralGCM for nearer-term forecasts over weeks to months. The work will proceed along three fronts: connecting the atmosphere and ocean, improving the response to a warming world, and making the tools easier to use and extend, including improved performance on widely used GPUs.

Nov 4Wed
  1. Fireworks AI Blog58

    Fireworks unveils new products at Forge to help companies build specialized intelligence

    At its first annual Forge conference, Fireworks launched products aimed at helping companies build AI around their own data and tasks, including Ultra-fast inference on NVIDIA Blackwell GPUs reaching up to 1,500 tokens per second, a half-price Background Mode, and multi-day prompt caching. It also introduced Fireworks Nexus for bringing open models into existing coding and coworking tools, No Eval and Specialized Model Inference for training and serving custom models, and Fireworks Sovereign, now in private preview, which runs its inference engine inside customers' own cloud or data centers. Fireworks says it already manages over 25,000 GPUs in customer environments.

  2. Databricks Blog34

    Declare and optimize with UNIQUE constraints on Databricks

    Databricks introduces UNIQUE constraints in Public Preview for Unity Catalog Delta tables in Databricks SQL and Runtime 18.2+, allowing tables to record multiple uniqueness guarantees beyond a single non-nullable primary key, with nullable columns and foreign key references supported. Constraints are informational and unenforced by default; after validation, marking them RELY lets the query optimizer eliminate eligible joins and DISTINCT aggregations, with benchmarks showing 7.9x and 12.0x speedups on TPC-DS SF10000.

  3. Hugging Face Blog55

    StateBench: Testing how language models track state changes over long documents

    Researchers introduce StateBench, a benchmark with four difficulty levels of long-context state tracking, from immutable assignment to general finite problems involving assignment by reference. In tests, open pretrained models and frontier models (Gemini 3.8 Flash, Claude Opus 5.5, GPT-6 Sol and Luna) degraded on harder levels as inputs grew, while 275M-parameter Gated DeltaNet and hybrid models trained directly on the tasks generalized far beyond training length, though transformers never learned the hardest level. The work is released with a tech report, data, and code, and notes open questions such as fixed variable counts and the exact hierarchy placement of constrained Gated DeltaNet variants.

  4. Meta Engineering34

    F3 Lightning: How Meta Accelerates Ads ML Features to Production at Scale

    Meta introduces F3 Lightning, an infrastructure layer on top of its Facebook Feature Framework (F3) that decouples feature releases from shared execution so each of the 20,000+ features in an ads ranking request can be compiled, validated, deployed, and rolled back independently. To preserve F3's efficiency, it uses stable computation identifiers, a single shared execution plan, one shared memory buffer per request, and dense index mapping for direct array access. Meta reports comparable extraction features reach production about 3.6x faster than prior F3 with no increase in infrastructure compute cost.

  5. Allen Institute for AI Blog55

    StateBench: Testing how language models track state changes over long documents

    Ai2 introduces StateBench, a long-context state-tracking benchmark with four difficulty levels (r-trivial, aperiodic, soluble, and general finite) grounded in Krohn–Rhodes theory. Testing open pretrained models and frontier models including Gemini 3.8 Flash, Claude Opus 5.5, and GPT-6 Sol and Luna, they found accuracy fell sharply on soluble and general finite problems as inputs grew longer, dropping to 13% or lower on the longest general finite problems. In from-scratch training of 275-million-parameter models, Gated DeltaNet and hybrid models generalized almost perfectly on the first three levels out to tens of thousands of assignments, while transformers never learned the hardest level.