DataExa 3.0 shifts the search budget toward selected evidence
The October 19 release expands the index and makes Dynamic Highlights generally available; shorter context still needs completeness checks.
Data
The datasets, collection systems and curation methods that determine what a model can learn.
Datasets · Collection & data rigs · Synthetic data · Curation & labeling
13 articles
Data
DataLinkup’s Aleph Alpha case study puts search inside the data boundary
The October 6 customer account describes EU-hosted search; an organization still needs to examine the complete path of each request.
DataHugging Face targets the text-normalization work behind BERT tokenization
The October 15 engineering report describes large throughput gains, with changes still at pull-request stage in that account.
DataDatabricks moves document parsing into interactive application workflows
Synchronous APIs shorten the route from incoming file to review, while extracted fields still need source checks.
Hover to animateDataNVIDIA expands Indic training data, but translation is only part of language coverage
NVIDIA and AI4Bharat announced a roughly 2.8-trillion-token Indic corpus and open generation pipeline; the corpus is described as releasing soon.
DataServiceNow turns agent failures into tasks that can actually be checked
AutoSynthData generates executable enterprise tasks from observed capability gaps, with positive and negative verification before training.
Hover to animateDataApptronik’s Robot Park makes the fleet part of the training system
Apollo 2’s wheeled and walking configurations collect different physical experience for the Google DeepMind collaboration.
Hover to animateDataCohere separates the cost of indexing from the cost of asking
Embed 5 Pro and Fast share an embedding space, allowing a corpus to use one model and live queries another. The compatibility has conditions and a measurable quality cost.
Hover to animateDataMecka: Human videos are valuable only if robots can learn from them
Human demonstrations could reduce the experience robots must collect themselves. Buyers need proof that the data transfers to their robot, task and environment.
Hover to animateDataTwelveLabs: AI video labels are useful when they save expert review
Pegasus 1.6 targets first-person footage used in physical-AI datasets. Faster annotation helps only if important actions and failures survive human review.
Hover to animateDataSnorkel AI: A flawed reward can teach an agent the wrong lesson
Training environments turn expert judgment into repeated lessons. Weak success checks can teach an agent to produce the right result through the wrong process.
Hover to animateDataDatology: Better data could make a smaller model the better choice
Curation Studio makes dataset selection a model-budget decision. The company’s reported compute saving is specific to its experiments; customers need a full-cost comparison.
Hover to animateDataNVIDIA: Define good synthetic data before generating more of it
NeMo Data Designer makes fields, dependencies and checks explicit. The useful asset is a reproducible training-data recipe, not the number of generated rows.
No articles in this topic yet
Explore the other topics in this desk.
Explore the research library for paper-based analysis.