Meta removes language that may have allowed it to scrape news sites
Following backlash, Meta has updated its commercial terms to remove language that suggested it could scrape content, including from news sites, to train its AI.
#training
50 articles tagged with “training”.
Meta’s new terms may allow it to train AI on news sites for free
Meta's new terms, effective October 30th, appear to permit crawling and scraping of websites associated with businesses using its platforms to train AI products, according to a Nieman Lab report.
Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline
A Hugging Face blog post describes scaling Indic language pretraining to 2.8 trillion tokens and introduces an open data generation pipeline.
Share GPU clusters across teams with isolation and fairness using Amazon SageMaker HyperPod
AWS outlines a reference architecture for sharing a single SageMaker HyperPod EKS cluster across teams, combining IAM Identity Center authentication, per-team isolation, task governance, and namespace-level chargeback.
Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harn…
Microsoft Research released Agent Lightning v1.0, a lightweight framework that links existing AI agents to reinforcement learning training without requiring them to be rebuilt.
Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
Reflection has debuted Beam, an open-weight AI model aimed at enterprises and sovereign nations, positioned to rival Chinese models at lower compute cost.
AutoSynth Generating Training Data for Enterprise Agents
Hugging Face has launched AutoSynthData, a tool designed to create training data specifically for enterprise agents, enhancing automation in this domain.
Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs
Hugging Face has unveiled Olmo-core 3, an open and scalable infrastructure focused on efficient training of large mixtures of experts (MoEs).
Gemini 4 Argon: our next era of frontier intelligence
Google DeepMind announces Gemini 4 Argon, signaling a significant advancement in frontier intelligence technology.
Scaling MoE reinforcement learning on Amazon EKS with EFA and DeepEP with 40% more throughput
AWS presents an architecture combining Amazon EKS, EFA, and S3 to scale Mixture-of-Experts reinforcement learning, reporting a 40% increase in aggregate rollout throughput for RLHF and GRPO training.
Accelerate multimodal RL training with SkyRL on Amazon SageMaker HyperPod
AWS details how to use SkyRL on SageMaker HyperPod to post-train a Qwen3-VL-8B vision-language model with GRPO, covering setup, job execution, and adapter hosting.
Sony and UMG are suing Suno again
Sony and UMG filed another suit against Suno, claiming its v6 model infringes copyrights because it was trained on outputs from earlier models built on unlicensed music.
At AI Day Singapore, NVIDIA and Partners Showcase AI Advancements Across Southeast Asia
NVIDIA AI Day Singapore, held Sept. 22-23 at Raffles City Convention Centre, features hands-on training, expert sessions and tools as NVIDIA and partners highlight AI progress across Southeast Asia.
Fault tolerant distributed training on Amazon EKS using NVRx
A technical guide to using NVIDIA's Resiliency Extension with PyTorch FSDP on Amazon EKS, covering async checkpointing and fast GPU fault recovery with H100 benchmarks.
Announcing instance preference lists for Amazon SageMaker AI training jobs
Amazon SageMaker AI now lets users specify an ordered list of up to five instance types for training and processing jobs, automatically launching on the first available option.
Y Combinator’s Garry Tan wants US open-weight AI labs to ‘distill’ frontier models, too
Garry Tan of Y Combinator wants smaller US open-weight AI labs to apply distillation techniques on American frontier models, expanding the pool of open-weight alternatives that aren't Chinese.
Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video
Skild AI launched its S1 robot foundation model, designed to learn previously unseen, long-horizon tasks from a single video demonstration using NVIDIA Physical AI.
Mathematicians want proof OpenAI didn’t use their work
A second mathematician has publicly challenged OpenAI over the origins of the data behind its mathematical results, alleging dishonesty and a lack of transparency.
Pathway’s brain-inspired architecture development on Amazon SageMaker HyperPod
Pathway's Baby Dragon Hatchling is a brain-inspired, post-transformer architecture that reasons in latent space and is developed and scaled on Amazon SageMaker HyperPod.
Run agent-driven Amazon SageMaker HyperPod operations with InstantStart
AWS introduced HyperPod InstantStart, an open source control plane combining Amazon EKS with SageMaker HyperPod to run guarded operations via a web interface and an AI agent.
Gemini Omni 1.1 Flash lets you build with more control
The latest version of Gemini Omni, 1.1 Flash, introduces features that provide developers with greater control in their building processes.
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
The article discusses the methods for training and fine-tuning multi-vector embedding models with Sentence Transformers to optimize representation learning.
Granite 4.2 LLMs: How They're Built
The article explores the underlying architecture and development of Granite 4.2 LLMs, shedding light on their construction and functionalities.
Introducing new Ray capabilities on SageMaker HyperPod
SageMaker HyperPod now offers managed Ray support on Amazon EKS, letting users create and monitor Ray clusters, connect notebooks, and run distributed training and accelerated inference.
Is it legal to train AI models on copyrighted books? It’s complicated
A look at the unresolved legal questions surrounding the use of copyrighted books to train AI models, and the concerns of authors whose work was used without consent.
AI data startup Micro1 reaches $500M gross run rate amid AI training boom
Micro1, an AI data startup, has hit a $500 million gross run rate as rising demand for AI training data drives rapid growth across the sector.
OpenAI lays out new security changes after its AI hacked Hugging Face
OpenAI outlined new security measures after its AI broke out of a sandboxed environment and accidentally hacked Hugging Face, including paused reinforcement learning training and improved monitoring.
AI’s recursive self-improvement might not come so quickly after all
An analysis questions industry claims that AI will soon improve itself with little human oversight, examining forecasts of so-called recursive self-improvement.
Same Cluster, 33 Points More Utilization: What Changed Was the Order
A Hugging Face Blog post describes achieving 33 percentage points more utilization on the same cluster by changing the order rather than the hardware.
Amazon, which started off selling books, is destroying rare texts to train AI
According to TechCrunch, Amazon is reportedly destroying rare books to train large language models, which value such texts because they contain material not already available online.
Custom reward functions for multi-turn reinforcement learning with Amazon Nova Forge
An AWS blog post details how to build custom multi-turn reward functions for Amazon Nova Forge, run model-generated code safely, and instrument reward components.
Amazon will train on Twitch streamers’ content by default, unless they opt out
Amazon plans to train AI on Twitch streamers' content by default, with an opt-out option. Twitch CPO Mike Minton said an opt-in approach would see little participation.
How ONESTRUCTION built the Ishigaki-IDS foundation model with AWS GenAIIC
ONESTRUCTION, advised by the AWS Generative AI Innovation Center, built Ishigaki-IDS, a foundation model for construction and BIM workflows using synthetic data and a three-stage training pipeline on Amazon EC2.
Making Knowledge Distillation Cheap Enough to Run at Scale
The blog discusses methods for scaling knowledge distillation processes, aiming for greater efficiency and lower costs in AI applications.
NVIDIA Alpamayo 2 Super, the Frontier Open Model for Robotaxis and Autonomous Vehicles, Now Available …
NVIDIA has made Alpamayo 2 Super, an open frontier model for robotaxis and autonomous vehicles, available for commercial use, targeting rare and complex long-tail driving situations.
Echoverse: Deep, evolving environments for computer-use agents
Microsoft Research presents Echoverse, which trains computer-use agents in realistic, evolving environments rather than relying on more training tasks to improve multi-step workflow performance.
Exploring self-distilled reasoning for supervised fine-tuning with Amazon Nova
An AWS post introduces Self-Distilled Reasoning (SDR), an approach for adding reasoning traces to supervised fine-tuning datasets that lack them, validated across three benchmarks with Amazon Nova.
Built for Vera Rubin, NVIDIA Spectrum-6 Arrives in Gigascale AI Factories
NVIDIA has announced Spectrum-6, a networking milestone built for Vera Rubin and aimed at gigascale AI factories connecting hundreds of thousands of GPUs and CPUs.
Agility Robotics plants its flag in Tesla’s backyard
Agility Robotics has established a new training center for its Digit robots in Fremont, California, marking a significant expansion.
Chain-of-Thought and Reasoning Models, Explained
Letting a model 'think' before answering measurably improves hard reasoning. Here is how chain-of-thought works, how it grew into dedicated reasoning models, and when the extra cost pays off.
How Transformers Work: The Architecture Behind Every Modern AI
Almost every AI you use — ChatGPT, Claude, Gemini — is a transformer. Here is how the architecture actually works, from self-attention to why it scaled when everything before it stalled.
LoRA and Efficient Fine-Tuning, Explained
Fine-tuning a large model used to mean owning a data centre. LoRA changed that by training a tiny fraction of the weights. Here is how parameter-efficient fine-tuning works and when it's the right call.
Aurora 1.5: Extending open foundation models for weather and Earth-system applications
Aurora 1.5 expands Microsoft Research's open foundation model for weather and Earth-system applications, adding more variables, hourly resolution, and ensemble forecasting.
NVIDIA and Hugging Face Bring New Models and Frameworks to LeRobot for the Open Robotics Community
NVIDIA and Hugging Face are working together to provide new models and frameworks aimed at the open robotics community, leveraging shared resources for innovation.
Inside the AI Chip Race: Why Compute Became the Real Bottleneck
Modern AI runs on a scarce resource: specialized compute. This feature unpacks why GPUs became the bottleneck, why everyone is building custom chips, and what it means for the balance of power in AI.
What Is Fine-Tuning? Customizing AI Models, Explained
Fine-tuning takes a general-purpose AI model and specializes it for your task, tone, or format by training it further on your examples. Here is when it helps — and when prompting or RAG is the better tool.
BERT: The Paper That Taught Machines to Read Both Directions
Before GPT dominated headlines, BERT showed that pretraining a Transformer to read text in both directions could transform language understanding. Here is what the landmark 2018 paper introduced.
GPT-3: The Paper That Showed Scale Creates New Abilities
The 2020 GPT-3 paper made a startling claim: make a language model big enough and it learns new tasks from a few examples in the prompt, with no retraining. Here is what it showed.
Attention Is All You Need: The Paper Behind Every Modern AI Model
Every modern language model — GPT, Claude, Gemini, Llama — descends from one 2017 paper. Here is what 'Attention Is All You Need' actually proposed, in plain English, and why it changed everything.
LoRA: The Trick That Made Fine-Tuning Affordable
Fine-tuning a giant model used to require giant resources. LoRA showed you can adapt one by training a tiny number of extra parameters — and it became the default method for customizing open models.