Skip to content
#Data / training
November 11Wednesday
  1. Latent Space62

    Mercor raises $750M Series D at $20B valuation on $3B annualized revenue

    Mercor raised a $750M Series D at a $20B valuation after reaching $3B in annualized revenue, up from $1B in February and $2B in June. The company reports 5 million experts and frames its market as the 800-plus US occupations whose workflows models have yet to be evaluated against. Mercor also says it already spends 3x as much on inference as on employee salaries, and argues that if the ratio holds across the economy it points to double-digit GDP growth over the next five years.

  2. Hugging Face Blog41

    Pooling-Aware Training for Multi-Vector Embedding Models with Sentence Transformers

    Hugging Face shows how pooling-aware training lifts retrieval quality for multi-vector embedding models at a fixed storage budget. With pool_factor=4, a normally trained model keeps 94.6% of its unpooled NanoBEIR nDCG@10 while cutting document vectors 4x, and pooling-aware training raises that retention to 97.8%. The post walks through HierarchicalTokenPooling, budget calibration and a reproducible Sentence Transformers example.

  3. inclusionAI Models40

    inclusionAI Releases ZoomGain-9B and ZoomGain-4B Models with ZoomGainBench and RL Training Data

    inclusionAI released the ZoomGain-9B and ZoomGain-4B vision-language model checkpoints, alongside the ZoomGainBench evaluation benchmark and ZoomGain RL training data. ZoomGain mines global-to-zoom perception gaps to build fine-grained visual supervision, and ZoomGain-9B reached superior performance on multiple fine-grained and general perception benchmarks. ZoomGainBench contains 1,000 questions over 941 images, while the RL dataset holds 9,000 training records and 7,713 source images.

  4. inclusionAI Models38

    inclusionAI Releases ZoomGain-4B and ZoomGain-9B with ZoomGainBench and RL Training Data

    inclusionAI released the arXiv paper, ZoomGainBench evaluation benchmark, RL training dataset, and ZoomGain-4B/ZoomGain-9B model checkpoints. The best of 28 models scored 61.08 on ZoomGainBench, while ZoomGain-9B reached superior performance on multiple fine-grained and general perception benchmarks. ZoomGain mines global-to-zoom perception gaps to build fine-grained visual supervision for vision-language models.

  5. Microsoft Research75

    Microsoft researchers describe Lazarus, a backdoor that reprograms dormant neurons in AI agents

    Microsoft researchers have described Lazarus, a backdoor technique that reprograms fewer than 10 dormant neurons in an LLM so an agent built on it emits attacker commands when a hidden trigger appears, with no retraining required. The modified model still passes standard evaluations and behaves normally until the trigger, which can be as simple as a date change, and the agent then runs the downloaded code with whatever access it was given. The authors say detection requires diffing against the original weights, and Microsoft Foundry already ties open-weight models to their publishers and scans their containers, while red-teaming harnesses, agent-based weight scanning and signed models with tamper-evident provenance are in development.

    Why it matters: Microsoft researchers show that reprogramming fewer than 10 dormant neurons can backdoor an LLM agent without retraining, which makes weight provenance and integrity a practical security requirement.

  6. Claude YouTube72

    Anthropic launches Claude Whiteboards, Sheets and Data for team collaboration

    Anthropic introduced Claude Whiteboards and Claude Sheets in beta, and made Claude Dashboards generally available as Claude Data. Whiteboards turns a spoken plan into a board with a timeline, owners and dates that teams can edit in real time, while Sheets builds spreadsheets with tabs, formulas, formatting and drop-downs and outlines each edit. Claude Data lets data teams certify queries and generate data stories or narrated videos from metrics, with live sharing showing up-to-date numbers.

    Why it matters: Claude adds collaborative whiteboards, spreadsheets and shared data dashboards, extending its workspace from chat into team documents.

November 10Tuesday
  1. Amazon Science62

    Amazon releases Mitra-v2 tabular foundation model with scaled synthetic pretraining

    Amazon introduced Mitra-v2, a second-generation tabular foundation model for classification and regression, released with fine-tuned models and inference and fine-tuning code under the Apache-2.0 license on Hugging Face. It is pretrained on synthetic tables with 10 times as many rows and roughly three times as many columns as Mitra-v1, using a new hybrid SCM prior and hybrid optimizer while keeping the Transformer architecture unchanged. On TabArena, a public benchmark of 51 real-world datasets, Mitra-v2 leads the single-model field, beats Mitra-v1 on 32 of the 33 datasets both run natively, and finishes all 51 within the one-hour budget; fine-tuned Mitra-v2 also ranks first on ScoringBench.

    Why it matters: Mitra-v2 shows that widening synthetic pretraining data, rather than deepening the network, lifts a tabular foundation model to the top of TabArena's single-model field.

  2. Recode China AI38

    Ulanqab, a Remote Chinese City, Becomes an AI Boomtown and a Magnet for Young Workers

    Ulanqab, a once little-known city in Inner Mongolia, has become an emerging AI computing hub as data centers spread across its grasslands, drawing young workers from across China. Kuaishou began operating its first self-built data center there in 2023, and DeepSeek advertised senior operations and maintenance roles paying up to 30,000 yuan ($4,200) a month this April. The labor market is now saturating, with employers raising hiring standards and some considering only bachelor's degree holders.

  3. Ars Technica · AI38

    Ukraine's Fiber-Optic Drones Squeezed by AI Data Center Demand

    Ukraine's fiber-optic drones, immune to battlefield jamming, now face higher costs as importing fiber material from China competes with global demand from AI data centers for the same limited supply. Ukrainian drone makers are adapting, with finished spools holding enough cable for drones to regularly operate at ranges of dozens of miles, and up to 62 miles in extreme cases.

  4. Apple Machine Learning Research34

    Weighted State Souping for Efficient Long-Context Retrieval

    Apple researchers and Harvard collaborators propose weighted state souping, a method that learns query-dependent weights for recombining cached SSM state encodings, improving long-context performance on single- and multi-hop document QA, needle-in-a-haystack retrieval, variable tracking, and tool use. A sparse variant achieves near-constant encoding cost with respect to context length while preserving or sometimes improving performance. The paper was accepted at the PALM workshop at NeurIPS 2026.

  5. Apple Machine Learning Research38

    Apple Study: Gating Mechanism Explains Why SSMs Memorize Instead of Learning In-Context

    An Apple and Harvard study shows that the gating mechanism in State Space Models (SSMs) causes them to first learn an in-weights "memorization" solution, delaying or even preventing convergence to correct in-context learning. This occurs even when the architecture and memory capacity impose no fundamental limits. Gating, however, often improves generalization to long sequence lengths.

  6. Apple Machine Learning Research31

    Apple study evaluates LLM judges for personalized writing style assessment

    Apple researchers tested how well LLM judges assess whether generated text matches a user's preferred writing style, generating text with four LLMs across three long-form writing tasks and comparing judge annotations against human pairwise labels. Judge quality correlated only weakly with general LLM ability on MMLU (Pearson's ρ ≤ 0.2) and varied by writing task and evaluation criteria, and paired comparisons proved more consistent than independent scoring. The strongest LLMs did not most reliably reproduce human evaluations, possibly due to exaggerated sensitivity to subtle details.

  7. Google Research25

    Google's Language Explorer Visualizes the World's Linguistic Diversity

    Google has launched Language Explorer, an interactive web platform that lets researchers, educators, and language enthusiasts discover and visualize linguistic diversity across thousands of languages. Built as part of Google's 1,000 Languages Initiative and based on the open-source LinguaMeta dataset, it is featured in a new exhibition at the Museum of Communication Berlin opening this week. Google Translate now supports nearly 250 languages, while Gboard covers 900+ language varieties.

  8. Latent Space58

    What Frontier Labs Buy: A Data Deep Dive with Handshake AI's Garrett Lord and Team

    Handshake's AI training business went from launch to $1 billion in gross annualized revenue in roughly a year, and this Latent Space deep dive with Garrett Lord and his team walks through the reinforcement learning environments frontier labs buy. In a finance demo, an agent working a simulated bank's email and deal files carries a goodwill figure into Python and subtracts it, a mistake that shows where expert judgment still beats the model. The team also presents VIALS, its benchmark for interpreting scientific images, and traces an agent's tool calls through safety simulations including CAREBench.

  9. Goodfire Research62

    Goodfire studies Qwen3.6-35B-A3B neurons as an interpretable basis

    Goodfire Research published a study arguing that individual MLP neurons in Qwen3.6-35B-A3B are more interpretable than assumed, based on sparsification, attribution tools and research agents. Ablating the 92.2% weakest MoE neuron activations per forward pass raised language model loss by only 2.9%, and 69% of neuron poles had a single mode responsible for at least 90% of their causal effect. Natural-language neuron descriptions let a judge model predict firing with 95% accuracy against random text and recover 73% of ablation damage through simulation.

    Why it matters: Goodfire's study of Qwen3.6-35B-A3B neurons argues that strong activations alone carry most of the model's computation, offering a concrete alternative to feature-dictionary interpretability.

November 9Monday
  1. Hugging Face Blog38

    NVIDIA and Elith Release Nemotron-Personas-Finance-Japan Synthetic Dataset

    Elith and NVIDIA released Nemotron-Personas-Finance-Japan, an open synthetic persona dataset for Japan's financial sector built on official government statistics. It contains 9M personas (1M records × 9 personas) across 50 fields and ~1.5B tokens, generated with NeMo SDG-PGMs and NeMo Data Designer, and is available under CC BY 4.0 for commercial use.

  2. Exponential View22

    AI-exposed workers took a month longer to find new jobs in 2025

    Workers who lost jobs in AI-exposed roles in 2025 took about a month longer than peers in other roles to find new employment, according to Exponential View's data roundup. Anthropic paid $3.4 billion to take over a supplier's two-year-old order for 1.7GW of gas generators, the largest known power-equipment purchase by an AI lab. Apple's memory inventory grew 2.5x year on year, helping shield it from memory cost increases.

  3. MIT Technology Review · AI12

    Lubrizol's AI-Driven R&D Strategy in Specialty Chemicals

    Lubrizol is building a digital foundation that connects business processes, modern data platforms, and cloud infrastructure to speed up R&D for specialty chemicals. IT vice president Girish Rao says the aim is to give scientists better tools, not replace scientific judgment, with "fusion teams" of scientists, data scientists, and IT staff starting from small pilots to build trust. Cloud access also supports sustainability work such as modeling starch and corn-based material alternatives.

  4. inclusionAI Repositories38

    ZoomGain: Mining Global-to-Zoom Perception Gaps for Fine-Grained Visual Supervision

    ZoomGain constructs questions about a full image from visual details that only become clear when zooming in, using this global-to-zoom perception gap as supervision for vision-language models. The release includes the arXiv paper, the ZoomGainBench benchmark (1,000 questions over 941 images covering attribute recognition, counting, and OCR), 9,000 RL training records over 7,713 source images, and the ZoomGain-4B and ZoomGain-9B checkpoints. The best of 28 models scored 61.08 on ZoomGainBench, while ZoomGain-9B reached superior performance on multiple fine-grained and general perception benchmarks.

  5. Preferred Networks News26

    PFN's AI Writing Instruction System Ronjiro Wins AI Special Award (Learning Support) at 23rd Japan e-Learning Grand Prize

    Preferred Networks' AI writing instruction system Ronjiro has won the AI Special Award (Learning Support) at the 23rd Japan e-Learning Grand Prize. Ronjiro uses generative AI including PFN's domestically developed PLaMo foundation model to support high school students' essay and composition learning, providing feedback on logical structure and improvement points in as little as one minute. It has been offered since March 2026 through YP Switch, a joint venture with the Yarakis Switch Group, and is now being introduced in some School IE classrooms nationwide.

November 8Sunday
  1. SemiAnalysis60

    SemiAnalysis visits Envision's 800MW Galaxy datacenter campus in Inner Mongolia

    SemiAnalysis toured Envision's Galaxy campus in Ulanqab, Inner Mongolia, planned for up to 800MW of IT capacity, with its first 120MW phase delivered within a year at about a fifth of US costs. The campus runs a live 5MW 800V DC pilot using solid-state transformers that convert 10kV AC straight to 800V DC for racks, and drops the central chiller by sharing one water loop between liquid- and air-cooled servers. Envision supplies over 60% of the first two phases' power from its own 420MW wind farms plus 4-hour battery storage, and the full campus is expected to be completed by the end of 2027.

  2. Artificial Analysis Articles38

    Artificial Analysis Decision Index v2 Released with New Decision Score and Private Benchmarks

    Artificial Analysis released Decision Index v2, with 45% of the Index now built on private datasets, up from 25% in v1. Drex Max v1.1 from Nace AI narrowly leads decision models at 73, within 1 point of GLiDE v1.2 and Mammoth 27B. The new Decision Score rewards both correctness and calibrated confidence, while AA-Paperclip and AA-Symmetry add private document and consistency tests.

November 6Friday
  2. Hugging Face Blog58

    Ai2 releases ToyBlocks, an open simulated block-manipulation world for robot learning

    Ai2 released ToyBlocks, an open simulation framework built around a robot arm and blocks on a tabletop, as the first of its FoundationWorlds for controlled robot learning and evaluation. The framework auto-generates skills and demonstrations via RL and motion planning in Isaac Lab, reaching up to 24x real-time on a single GPU, and ships with 85,336 long-horizon demonstrations covering nearly 1,600 hours and 144 million frames. In its experiments, ManiFlow's atomic-skill success was 80.0% in simulation and 81.1% on a real Franka FR3, while the Molmo 2 plus ManiFlow system succeeded in 55.5% of simulated and 50.0% of real-world episodes.

  3. MIT Technology Review · AI38

    I Watched a Tau Robotics Humanoid Robot Clean My Apartment

    MIT Technology Review tested Tau Robotics' humanoid cleaning service in San Francisco, where robots are remotely piloted by humans rather than fully autonomous. The robots collect training data for future autonomous models, with pricing at $60 per weekly cleaning with two robots or $100 for a one-time visit. The robot cleaned a kitchen and toilet but struggled with stairs and repeatedly stalled during the first visit.

  4. vLLM Blog77

    REINFORCE Zero: Kernel Alignment Enables Stable Async RL at Policy Lag 64, Extends to DeepSeek-V4.1 and Kimi K3.1

    NVIDIA Research and Inferact's REINFORCE Zero paper shows that aligning rollout and trainer kernels lets unclipped token-level importance sampling correct the remaining weight lag, training stably for 20,000 updates at policy lag 64 and reaching 2.40x the training-step throughput of synchronous training with aligned kernels. With standard kernels, the median rollout-to-learner KL rises 83.33x across the run and online AIME24 pass@1 falls from 34.27% to 13.96%, while a sequence gate stabilizes training but rejects 24.48% of sequences by the last 1,000 updates. The zero-mismatch setup now covers DeepSeek-V4.1-Flash, Kimi K3.1 on release day, and DeepSeek-V4-Flash and Kimi K3.1 on AMD MI355X, with code in Inferact/zero-mismatch-rl and NVIDIA-NeMo/labs-molt.

    Why it matters: NVIDIA Research and Inferact split the async RL rollout-learner gap into weight lag and kernel mismatch, showing that aligning kernels lets unclipped importance sampling hold stable training at policy lag 64.

  5. vLLM Blog55

    VeRL-Omni v0.3.0 released with async training and broader multimodal post-training

    VeRL-Omni v0.3.0 was released with asynchronous and decoupled RL execution paths, the V1 trainer as the default, and expanded post-training coverage across image, video, audio, and multimodal understanding. The release adds fully async Qwen3-Omni text and audio training, separate-async Omni and diffusion training with a replay buffer, and new recipes including MiniMax H3 with DiffusionNFT and MiniCPM-o 4.5 GSPO. In a MiniCPM-o 4.5 AVQA experiment on 4×H800 80 GB, separate-async 3+1 was about 9% slower per prompt than colocated 4+0 at natural response length, roughly tied at 512 tokens, and about 6.7% faster at 1024 tokens.

  6. Anthropic Research60

    Anthropic explains the components of Claude agent memory

    Anthropic outlined how it designs memory for Claude agents, defining agent memory as knowledge retained across sessions and separating it from weight changes, in-context learning and compaction. 5 matched it with memory. 7 wrote 413 files across 225 folders and earned one badge.

    Why it matters: Anthropic details how Claude's agent memory works, comparing models' note-taking on CL-Bench and Pokémon Red and showing how offline dreaming corrects and organizes stored notes.

November 5Thursday
  1. The Pragmatic Engineer22

    Scaling Slack and GitHub: Maude Lemaire on Backend Performance and Database Sharding

    Maude Lemaire, principal engineer and author of "Refactoring at Scale," shares an insider view of scaling Slack and GitHub on The Pragmatic Engineer podcast. She spent nearly eight years leading Slack's backend performance infrastructure team, then joined GitHub in 2024 to tackle database scaling, and most recently worked on performance at Cursor, including time to first token. She explains Slack's 23-custom-emoji limit to avoid thundering herds, how load testing 500,000 WebSocket connections the naive way would have cost $150,000 per run, and why GitHub's Active Record abstraction and scatter queries capped it at 16 database shards.

  2. Amazon Science18

    Amazon Scientist Hamed Karimi Wins 2026 ECML PKDD Test-of-Time Award

    Amazon senior applied scientist Hamed Karimi and coauthors Julie Nutini and Mark Schmidt received 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 this 1963 inequality guarantees linear convergence even when models lack a single best solution, unifying scattered optimization results and extending them to nonsmooth landscapes and algorithms such as coordinate, sign, and stochastic gradient descent. It has been cited over 2,000 times and appears in graduate textbooks.

  3. inclusionAI Repositories38

    ASpec: A High-Throughput Training Framework for Speculative Decoding Draft Models

    ASpec is a high-throughput training framework for speculative decoding draft models, built from the TorchSpec codebase with optimized attention, fused kernels and selective head/loss computation. On a Qwen3-8B 4K run, Draft-only training reaches 15,256.44 input tokens/s on one GPU, 1.96× TorchSpec, while the 8-GPU online steady window hits 77,721.37 tokens/s, 2.28× TorchSpec. It supports DSpark, DFlash, EAGLE3/4 and native MTP drafts, with Qwen3-8B and Ling tiny quickstarts.

  4. Apple Machine Learning Research38

    Apple's Learned Support Functions Speed Up RAG Retrieval

    Apple researchers presented "Amortizing Maximum Inner Product Search with Learned Support Functions" at ICML 2026, training small neural networks offline to route queries toward their best-matching keys faster. SupportNet learns the support function via input convex neural networks (ICNNs) to improve cluster routing, while KeyNet directly outputs the best-matching key vector, kept geometrically grounded by a consistency loss from Euler's theorem. Both are trained on the actual query distribution, trading an upfront amortization cost for faster, more accurate retrieval at inference time.

  5. Google Research47

    Google Research's Agile Deliberation helps humans define subjective AI categories

    Google Research introduced Agile Deliberation, a framework that turns subjective AI classifier creation into an iterative conversation where the AI helps users sharpen their own definitions. It splits the task into concept scoping and concept iteration, using vision-language models to surface borderline images and route feedback into an automatic prompt optimization loop. In 18 live 1.5-hour user sessions, the approach reduced user fatigue and anxiety and produced more consistent models that outperform standard baselines; the code is open-sourced on GitHub.

  6. Anthropic Research58

    Anthropic studies how researchers use Claude across scientific fields and tasks

    Anthropic analyzed 207,049 Claude and Claude Code research sessions from August 2 to September 30, 2026, using privacy-preserving classifiers, and found scientists use Claude most for data analysis and writing and least for data gathering and ideation. Data analysis accounts for 33.7% of Claude sessions and 48.1% of Claude Code sessions, against 27.6% of researcher time, while ideation takes 4.2% and 2.1% of sessions versus 7.1% of researcher time. The report estimates that wide adoption of current capabilities could roughly double research output in some fields, and full automation of non-physical work could lift fields such as computer science and statistics as much as 15-fold.

  7. Goodfire Research58

    Goodfire Research proposes mechanistic shaping to steer models via internal representations

    Goodfire Research proposes mechanistic shaping, a framework that adds penalties on a model's internal concept directions during training so it learns to rely on or express specific concepts rather than just maximize task reward. In chest X-ray experiments, an LLM agent used sparse autoencoder features to find tube-like shortcut directions, and reliance shaping beat both standard training and a balanced-data baseline.

November 4Wednesday
  1. Ai2 · Allen Institute for AI38

    Ai2 and Google collaborate on NeuralGCM+ climate models

    Ai2 has launched a climate modeling initiative backed by three-year funding from Google.org and technical support from Google Research to develop NeuralGCM+, an open-source model for faster, more reliable climate projections. The project extends Google Research's NeuralGCM, which combines physical equations for large-scale dynamics with AI for smaller-scale processes like cloud formation and rainfall, to simulate atmosphere and ocean together. Ai2 aims to cut regional temperature and rainfall errors over the past 50 years and release models and frameworks with documented workflows for outside researchers.

  2. SemiAnalysis65

    SemiAnalysis: SpaceX will become the world's largest AI cloud, with ~13GW targeted for 2027

    SemiAnalysis reports that SpaceX now has 3.6GW of datacenter capacity under construction across multiple sites and has raised its 2027 estimate to roughly 13GW, up from the 10GW figure it gave on its first earnings call. The firm says it has already identified more than 9GW of feasible capacity and argues SpaceX may be the fastest datacenter builder in the world, citing Colossus 2 build times of 91 days for 210MW and a new Mississippi site targeted at about 4 months. It also flags a likely datacenter at the TeraFab chip fab site in Texas and a BESS deployment at Colossus 2 approaching 6GWh.

  3. Meta Engineering Blog34

    Meta F3 Lightning Accelerates Ads ML Feature Releases ~3.6x

    Meta introduced 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+ ads ranking features is compiled, validated, deployed, and rolled back independently. Comparable extraction features now reach production ~3.6x faster than prior F3 with no increase in infrastructure compute cost.

  4. Anthropic Research66

    Anthropic biology lab reports first prokaryotic vault particle structure, solved with Claude

    Anthropic's biology research lab reports the first 3D structure of a vault particle found in prokaryotic cells, a phage protein called L05 built from 96 copies with a mass of about 9.5 megadaltons. Under the team's direction, Claude searched a database of about 15 billion predicted proteins, designed DNA constructs, optimized a size-exclusion chromatography isolation that cut purification from three days to about 90 minutes, and ran RELION cryo-EM processing that produced a first full model in 34 hours. The preprint has been submitted to bioRxiv, and the team says the vaults' biological function remains unknown.

    Why it matters: Anthropic's biology lab reports the first 3D structure of a vault particle from prokaryotes, with Claude handling database search, construct design, purification and cryo-EM processing.

November 3Tuesday
  1. Ai2 · Allen Institute for AI55

    Ai2 releases StateBench to test how language models track state changes across long documents

    Ai2 introduced StateBench, a benchmark with four difficulty levels for long-context state tracking, from immutable assignment to self-referential and cross-referential updates. In tests, Gemini 3.8 Flash, Claude Opus 5.5, and GPT-6 Sol and Luna handled the easier levels but dropped to 13% or lower on the longest general finite problems, while 275M-parameter Gated DeltaNet and hybrid models trained from scratch generalized almost perfectly on the first three levels and transformers never learned the hardest one.