#Data/Training

AIA native word-level transcription model outputs structured, timestamped arrays of word, spacing, and audio_event tokens directly from audio input, without a secondary forced-alignment pass. Audio events such as laughter or applause are tagged separately, which the source says helps with captioning, searchable archives, and highlight identification. The source notes Scribe's word-level transcription supports up to 5 independently transcribed channels.

AIPromptArmor reports that a malicious Skill can make Databricks Genie Code display a phishing modal and exfiltrate tenant data without human approval. The attack exploits Skills loaded from users' personal workspaces and a display interface that lacks egress controls, and Databricks, after disclosure on August 16, 2026, said users are responsible for ensuring uploaded Skills contain no malicious content.

AIApple and Stanford researchers built two open-ended probes using a Wizard of Oz technique so participants could train personalized machine learning systems on phenomena they defined themselves. In a week-long exploratory study, participants identified four sites where ontological boundaries were negotiated: the boundaries of a phenomenon, the subject as part of relations, signal versus noise, and the objectivity of data. The paper offers starting points for supporting boundary negotiation through design.

AIIndexTeam has published an official NVFP4 (W4A4) quantized version of its Index-Echo-S2TT-2B speech-to-text translation model on Hugging Face. Only the text LLM backbone is quantized, while the audio tower, connector, and speech-synthesis components remain in BF16. Perplexity rises 5.80%, from 4.8772 to 5.1599, on a fixed corpus, and full FP4 speedup requires an NVIDIA Blackwell GPU.

AIThe vLLM team integrated Helion, a PyTorch-native kernel DSL, into vLLM's linear backend, using per-shape autotuning to select among Standard GEMM, Split-K, and Swap-AB variants. On NVIDIA Hopper GPUs, the Helion backend outperformed the default CUTLASS and DeepGEMM backends across the evaluated models, with more than 10% throughput gains for some workloads. The work focuses on FP8 and INT8 quantized GEMM.

AIEpoch AI estimates that compute built from projected 2025 to 2027 high-bandwidth memory shipments could support tens to hundreds of millions of frontier AI agents, or billions of cheaper ones. Running nonstop, the top-tier agents would match the working hours of 140 million to 700 million full-time employees, and the central DeepSeek V4 Pro estimate of about 1.9 billion agents would match 8 billion workers.

Why it matters: The estimate converts memory shipments into agent capacity and revenue ranges, showing how hardware supply could translate into labor and sales if demand keeps up.

AIAi2 released AstaBrief 8B, an open-weights model that turns a research question and retrieved literature excerpts into a cited report, along with its training data. The model runs as Fast mode in Asta, averaging 51.1 seconds per report versus 178.5 seconds for Thinking mode, about 3.5x faster. The post also describes filtering synthetic training data by citation density and building DPO pairs judged by two models that agreed.

Why it matters: The post explains how supervised fine-tuning, preference data, and citation-density filtering were used to build a cited-report model, which is useful for teams training their own models.

AIGoogle announces a next-generation federated learning system that uses Trusted Execution Environments to provide verifiable, auditable data anonymization. The system publishes access policies to a public transparency log and is deployed in Gboard, which has launched English and Japanese next-word prediction models with stronger privacy guarantees and improved accuracy. Training time has also sped up significantly because computation moved to the server and is parallelized across many machines.

Why it matters: The post shows how Trusted Execution Environments make federated learning's privacy claims externally verifiable, rather than relying on trust in the server operator.

AIAi2 released AstaBrief 8B, a model that turns a research question and retrieved literature excerpts into a cited report, along with its training data. In Asta's Generate a report feature, Fast mode averages 51.1 seconds per report versus 178.5 seconds for Thinking mode, about 3.5x faster. The model is built on Qwen3-8B with supervised fine-tuning and DPO, and institutions can run its open weights on their own infrastructure.

Why it matters: The post explains the data filtering and one-pass generation choices behind a fast open-weights report model, showing what worked and what did not.

AIServiceNow CoreAI introduced AutoSynthData, which uses a target model's failures and a stronger teacher's successes to generate and validate new agent training tasks. In EnterpriseOps Gym experiments, the Hybrid domain produced 2,000 samples and raised Gemma-4-26B-A4B-it mean Pass@1 by 7.2 percentage points, while the ITSM domain produced 1,994 samples and raised it from 18.77% to 27.18%.

Why it matters: The post shows how failure analysis, teacher demonstrations, and verifier checks combine into a repeatable pipeline for generating targeted agent training data.

AIResearchers Maureen de Seyssel, Jie Chi, and Zakaria Aldeneh found that strengthening language discrimination during pretraining reduces the performance gap between multilingual and monolingual HuBERT speech models. In a controlled English/French setting, phone-ABX error fell from 11.6% to 10.4%, close to the monolingual 10.8%, while lexical sWUGGY scores rose from 52.1% to 56.7%. The gains were largest when language discrimination was introduced in the first training iteration.

AIEpoch AI estimates that high-bandwidth memory shipped in 2025–27 could eventually support about 30–170 million concurrent frontier-model agents once fully deployed and allocated. Using DeepSeek V4 Pro serving benchmarks, the estimate rises to about 1.9 billion concurrent agents. The authors compare the implied API-equivalent spending of $2.6–5.3 trillion per year with projected developer revenue of roughly $1 trillion by end-2027, suggesting demand may lag supply.

Why it matters: The analysis converts HBM shipment data into concurrent agent capacity and compares it with projected API revenue, showing where compute buildout may outpace demand.

AIApple Machine Learning Research reports that discrete diffusion steps match the training distribution only when simultaneously written token positions are conditionally independent given already-fixed tokens. The authors show that per-position distributions cannot determine such dependence, and on the synthetic ScanAndAdd task, confidence-ranked groups of two or more positions were dependent and produced a generated distribution 29 times the sampling-noise floor in total variation.

AIGoogle's prototype satellite for Project Suncatcher, built with Planet, launched into orbit on the Transporter-18 rideshare mission with SpaceX. The team confirmed contact and says the satellite is operating as expected. Over the coming weeks, it will gather in-orbit data on how Google's TPUs handle spaceflight stress, radiation, and thermal extremes, and a peer-reviewed paper detailing the research is available in Joule.

AIGoodfire Research developed sequence-aware monitors using protein language model embeddings to flag concerning biological sequences in dual-use AI agent tasks. On a custom benchmark, the monitors outperformed frontier model safeguards with fewer refusals on benign requests, and they held up better against paraphrasing and fragmentation attacks. The paraphrase results rely on in-silico estimates and do not establish whether the redesigned proteins keep biological activity, and the monitors run in milliseconds per sequence.

Why it matters: The post gives a concrete benchmark setup and fragmentation results, showing how sequence embeddings can separate dual-use biology requests that task-based safeguards handle poorly.

AICloudflare's AI Search is now generally available, adding native image embeddings for visual retrieval, OCR for scanned PDFs, and a 10 MiB file limit up from 4 MiB. Billing begins November 1, 2026, charged per ingested token, stored GB-month, and query, with a free monthly allotment on all Workers plans.

AIAi2 released Olmo-core 3, an open training framework redesigned to scale mixture-of-experts models into the trillion-parameter range. In one benchmark, expert count rose from 8 to 128 with about 3.2B active parameters per token, total capacity grew from 4.6B to 47B, and throughput fell by less than 5%. The framework is fully open, so researchers can train their own MoEs and experiment with routing and parallelism.

Why it matters: The release documents concrete MoE scaling results and reported failure modes, useful for teams weighing training-stack tradeoffs before adopting an open framework.

AIPhysicist Matthew Schwartz describes building BootLoops, an open-source harness for exact quantitative calculations, after choosing problems suited to Claude's strengths. He reports that Claude solved long-standing integrals and found connections across ecology, population genetics, economics, and linguistics, with domain experts steering results toward questions those fields care about. The post states that the approach required constant human oversight, since Claude often overstated results and misjudged time.

Why it matters: The guest post explains why scientists often find current AI tools frustrating and offers a method for finding problems where AI and researchers match, backed by concrete projects.

AIMastra platform now lets users attach a VPC-isolated Postgres database to any environment, restricting access to resources on the same network. The database cannot be reached from outside the network, so psql connections from external clients return an error. VPC Postgres joins Turso and Neon as managed database options, with MongoDB and Redis coming soon; it requires mastra@1.32.0 or later.

AIApple researchers introduced RLTL;DR, a reinforcement learning method in which an agent writes its own one-line insight after each failed attempt and learns to map tasks to those insights. On challenging tool-calling and coding datasets filtered to Pass@128 = 0, standard GRPO training of a Qwen 3.5 9B Thinking policy stayed at 0% to 1% Pass@1, while RLTL;DR reached 14–31% with insights in context and 12–13% without them at evaluation. A compact variant, SFTL;DR, trained on just 4k task-insight tuples recovered nearly the full performance of RLTL;DR.

AIMicrosoft Research intern-developed machine learning pipeline forecasts location-specific geomagnetic risk for 66,935 substations in the continental United States. It combines solar-wind observations, AE and Dst forecasts, geological conductivity and grid data to estimate risk 30 to 60 minutes ahead. The pipeline detected nearly 80% of major space-weather events during the evaluation period.

AIMicrosoft says its Circular Centers now operate eight facilities across North America, Europe, and Asia Pacific to decide the next life of decommissioned Azure hardware. Last year, Microsoft achieved a 92% reuse and recycling rate for decommissioned servers and components. Since 2014, Azure cores per rack have increased about 13-fold while power for the same task fell roughly 90%.

AINumerical differences between a rollout engine and a trainer can make reinforcement learning collapse even when algorithm and data stay identical. In a GLM 5.2 experiment, reward fell from about 0.9 to under 0.2 around step 20 without alignment, while aligned numerics kept reward stable over 25 steps. A Qwen3.5-MoE investigation traced a significant mismatch to how expert outputs were combined, and router replay alone was judged insufficient.

AIMicrosoft's Azure AI team argues that better models do not eliminate the need for a dedicated content extraction layer, since agents need trustworthy, structured, and auditable inputs. The post notes that building extraction directly on an LLM quickly demands chunking, layout parsing, grounding, normalization, and evaluation infrastructure. Microsoft positions Azure Document Intelligence and Azure Content Understanding in Foundry Tools as managed options for that layer.

AIMicrosoft has made SQL Server on Azure Local generally available for connected and disconnected deployments, letting organizations run SQL Server in their own datacenters and edge locations. Disconnected operations continue locally where external connectivity is restricted or unavailable. Eligible existing SQL Server licenses can be used, and Foundry Local on Azure Local, currently in preview, brings AI inference alongside SQL Server data.

AIMicrosoft Research introduced Quine, an experimental research system combining a multimodal world model of biology with an interactive harness that connects models, scientific tools, literature, and researchers. In a pancreatic cancer study with the Broad Institute, Quine prioritized compounds that shifted tumor cell states, and several top-ranked candidates were validated in wet-lab assays. Access is initially limited to the Quine Fellows program and select collaborations, and the system is intended for research use only, not clinical use.

Why it matters: The post shows how a multimodal biology world model is wired into a harness, grounded in one wet-lab cancer example and a limited fellows-program access path.

AIMicrosoft is bringing business context from Fabric IQ into Microsoft Copilot, with Fabric IQ in Copilot Chat and Cowork generally available and integration into the new Code experience coming soon through the Frontier program. Power BI is also gaining agentic app creation in Power BI Desktop, letting users generate applications from trusted semantic models and publish them to Microsoft Fabric.

AIMastra has added memory hooks that let developers monitor or alter an agent's observational memory cycles. Lifecycle hooks such as onObservationStart and onReflectionEnd report on each cycle, including token usage for spotting cost spikes, while transform hooks like beforeObservation and afterReflection can prune, remove, or redact memory data.

AIMicrosoft Research Asia – Singapore, opened July 24, 2025 as Microsoft's first Southeast Asian research lab, reports progress after its first year. The lab's work spans next-generation AI models and agentic systems, domain-specific AI for real-world impact, AI-native research practices, and ecosystem and talent development. Its healthcare collaborations on multimodal and agentic AI for clinical decision-making are being deployed through partnerships across Singapore's healthcare ecosystem.

AIEpoch AI estimates that the cheapest cost of reaching a given benchmark score has fallen about 13× per year over the past five years, faster than DNA sequencing, compute, lithium batteries, or electricity. Its example: a 75% GPQA Diamond score that cost about 30 cents per question with o3 in January 2025 cost $0.0004 per question with GPT-5.6 Luna under 18 months later. The authors caution that benchmarks are imperfect proxies for market prices, and the decline rate slows over time.

Why it matters: The source compares AI price declines with other transformative technologies using benchmark-based cost estimates, giving readers a measured sense of how fast cost per capability is falling.

AIXiaomi MiMo reports that tool-call repetition in MiMo-V2.6 reached over 0.05% of responses across agent harnesses, causing stalled agents and wasted context. The team traced the cause to an RL flooding penalty set at 32 calls per turn, which missed smaller excess behavior, and replaced the approach with a specialized teacher distilled via MOPD. Repetition rates for both Pro and Flash dropped substantially, at roughly $90,000 versus an estimated $2.31 million for the alternative fix.

Why it matters: The post traces an agent failure to a reward blind spot and compares the costs of two fixes, offering a transferable debugging method for RL-trained tool-calling models.

AIShanghai AI Lab's InternLM released Intern-Decision-4B, a multimodal structured decision model fine-tuned from Qwen3.5-4B, which returns answer distributions for multiple questions in one forward pass. On its benchmark table it scores an average of 90.02 with a Brier score of 0.347 and an ECE of 0.065, and per-query latency averages 44.16 ms on a single RTX 4090. The model is available with a Python DecisionEngine inference interface.

AIInternLM released Intern-Decision-0.8B, a multimodal structured decision model fine-tuned from Qwen3.5-0.8B that scores answers to multiple questions in one forward pass. The model reports a 79.38 average score and a 33.98 ms mean latency on a single RTX 4090, with 0.8B, 2B, and 4B sizes available. It is accessed through a Python DecisionEngine API that returns calibrated probabilities rather than generating free-form text.

AIGoogle Cloud has launched a managed reinforcement learning fine-tuning service (RLFT) that lets customers adapt Gemini models using a reward function they define instead of labeled answers. Users supply prompts and a reward function, while Google handles the RL infrastructure and proprietary model internals. The guide advises exhausting prompting and supervised fine-tuning first, and notes that RLFT suits tasks that are easy to score but hard to demonstrate.

AIReplit has acquired San Francisco-based Atta, bringing its business analysis and purpose-built charting into Replit. The first capability, interactive charts directly in Replit chat, is available today, and users can ask Replit Agent to visualize data or request a specific chart without writing SQL.

AIMicrosoft Foundry expands its agent platform with voice agents in public preview, long-running resilience for hosted agents, and tools for evaluating production agents. The post also says GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5 are now available in Foundry. Agent optimizer, Insights, and Rubric evaluator are described as tools for continuous improvement, with some reaching general availability later this month.

Why it matters: The post shows how Foundry combines model choice, voice agents, long-running resilience, and production evaluation into one agent workflow, with a customer example.

AIGoogle's Project Suncatcher will launch a prototype satellite on the Transporter-18 rideshare mission with SpaceX to test how its TPUs handle spaceflight. Initial ground tests showed the Trillium TPUs survived vibration and a radiation dose greater than a five-year space mission would deliver. Google says cooling with heat pipes and radiators and laser links between satellites in 2027 remain open engineering challenges.

Why it matters: The source reports concrete radiation, vibration, and cooling test results for TPUs, showing what space-based AI compute still has to solve.

AIGoodfire Research examines how language models track emotional dynamics across a story, sentence by sentence. The authors prompt Llama 3.1 8B to rate six emotions after each sentence, then harvest activations from each sentence's last token and fit a manifold to show stories tracing trajectories through it.

AIGoodfire Research examines how sparse autoencoder directions relate to curved manifolds in neural representations, identifying shattering, compact capture, and dilution as three ways lines can represent them. The team trained an autoencoder on synthetic data containing shapes such as donuts, spheres, and Möbius strips, and reports that real features in Llama 3.1 8B show dilution. It also describes an unsupervised pipeline that clusters features by firing patterns to surface manifolds in that model.

AILangChain launched LangSmith Fine-Tuning and smithtune, a CLI that turns LangSmith agent trajectories into fine-tuned models through dataset creation, training with Fireworks or Baseten, and evaluation in LangSmith. smithtune currently supports supervised fine-tuning, training models on recorded examples of good agent behavior by updating model weights. The tool lets teams train specialized models without building the data pipeline by hand.