✷ The archive · 11 labs · 3,433 dispatches
The labs
No more opening a dozen official blogs every morning. First-hand releases from OpenAI, Anthropic, DeepMind and the rest, each with its substance pulled out.
Hugging Face Private Hub Introduction
Hugging Face introduced the Private Hub (now Enterprise Hub), a secure, compliant environment for companies to collaborate on machine learning models, datasets, and spaces using a unified toolset.
Nyströmformer: Approximating self-attention in linear time and memory via the Nyström method
Nyströmformer reduces the time and memory complexity of standard self-attention from O(n^2) to O(n) by using the Nyström method to approximate the softmax matrix via sampled landmarks.
Hugging Face Recommendations for the U.S. National AI Research Resource
Hugging Face submitted a response to the White House and National Science Foundation advocating for the National AI Research Resource (NAIRR) to prioritize ethical expertise, standardized documentation, and accessible tooling to democratize AI research.
OpenAI Announces Efficient Training of Language Models for Fill-in-the-Middle
OpenAI released a new training method that enables autoregressive language models to perform fill‑in‑the‑middle tasks without sacrificing left‑to‑right generation quality, offering a simple and efficient way to add infilling capability.
Hugging Face Datasets Audio and Vision Documentation Update
Hugging Face has expanded the 🤗 Datasets library documentation to include dedicated guides and tools for audio and vision datasets, simplifying the process of loading and processing multi-modal data.
Faster Text Generation with TensorFlow and XLA
Hugging Face has enabled XLA compilation for text generation in the transformers library using TensorFlow, resulting in speedups of up to 100x and performance that often exceeds PyTorch.
A Hazard Analysis Framework for Code Synthesis Large Language Models
OpenAI has introduced a hazard analysis framework to identify and mitigate technical, social, political, and economic risks associated with the deployment of advanced code generation models like Codex.
Deploying TensorFlow Vision Models in Hugging Face with TF Serving
Hugging Face provides a guide on deploying TensorFlow vision models, such as the Vision Transformer (ViT), using TensorFlow Serving to expose them as REST or gRPC endpoints.
Advantage Actor Critic (A2C) Explained
Hugging Face explains Advantage Actor Critic (A2C), a hybrid reinforcement learning architecture that reduces variance in policy gradient estimation by combining policy-based and value-based methods.
OpenAI DALL·E beta release
OpenAI announced that DALL·E is now available in beta for 1 million invited users, offering free monthly credits and commercial usage rights.
Reducing Bias and Improving Safety in DALL·E 2
OpenAI has implemented a system-level technique for DALL·E 2 to increase the diversity of generated people and enhanced safety filters to prevent the creation of deceptive content.
Dynamic Adversarial Data Collection (DADC) for MNIST – Hugging Face tutorial
Hugging Face announced a step‑by‑step tutorial for Dynamic Adversarial Data Collection (DADC) on MNIST, showing how to gather human‑generated adversarial examples and retrain models to improve robustness.
DALL·E 2: Extending Creativity
OpenAI's DALL·E 2 research preview has been adopted by over 3,000 artists across 118 countries to integrate AI-generated imagery into diverse creative workflows ranging from medical support to high-end culinary arts.
BLOOM 176B Training Technology Overview
Hugging Face released a detailed breakdown of how the 176‑billion‑parameter BLOOM model was trained using 384 A100 GPUs, Megatron‑DeepSpeed 3D parallelism, and BF16 mixed‑precision, highlighting the hardware, software stack, and engineering challenges.
Building a Playlist Generator with Sentence Transformers
Hugging Face demonstrates how to create a semantic search-based playlist generator using Sentence Transformers for lyric embeddings and Gradio Blocks for the user interface.
BLOOM: The World's Largest Open Multilingual Language Model
Hugging Face and the BigScience project have released BLOOM, a 176-billion parameter open-access multilingual language model trained in complete transparency across 46 natural languages and 13 programming languages.
Anthropic Research: Language Models (Mostly) Know What They Know
Anthropic research demonstrates that large language models can effectively evaluate the validity of their own claims and predict their likelihood of knowing an answer, providing a path toward more honest AI.
Getting Started with Sentiment Analysis on Twitter
Hugging Face provides a guide on using the Inference API to perform automated sentiment analysis on Twitter data for both developers and non-coders.
Policy Gradient with PyTorch – Hugging Face Deep RL Class Unit 5
Hugging Face released a tutorial that walks through implementing the REINFORCE Monte‑Carlo policy‑gradient algorithm from scratch in PyTorch and evaluates it on CartPole‑v1, PixelCopter, and Pong.
Getting Started with Machine Learning using Sentence Transformers
Hugging Face provides a guide for beginners to start their first ML project using the Sentence Transformers library to create embeddings and perform semantic search.
DALL·E 2 Pre-training Mitigations
OpenAI implemented active learning for data filtering, a reweighting scheme to prevent bias amplification, and a clustering-based deduplication process to eliminate image regurgitation in DALL·E 2.
Hugging Face Evaluation on the Hub
Hugging Face has introduced Evaluation on the Hub, a no-code tool powered by AutoTrain that allows users to evaluate any model on any dataset using any metric directly on the Hugging Face Hub.
Accelerate Large Model Training using DeepSpeed
Hugging Face demonstrates how to use the Accelerate library to integrate DeepSpeed's ZeRO optimizations, enabling larger batch sizes and the training of models that would otherwise exceed GPU memory.
OpenAI Video PreTraining (VPT) for Minecraft
OpenAI introduced Video PreTraining (VPT), a semi-supervised imitation learning method that enables agents to learn complex behaviors in Minecraft by training on massive unlabeled video datasets of human play.
Getting Started With Embeddings – Hugging Face tutorial
Hugging Face announced a step‑by‑step guide for creating, hosting, and querying vector embeddings using the Sentence‑Transformers library and the Hugging Face Hub, demonstrating a searchable FAQ system.
Convert Transformers to ONNX with Hugging Face Optimum
Hugging Face provides three methods to convert Transformers models to ONNX, ranging from the low-level torch.onnx API to the high-level Optimum library.
OpenAI Evolution through Large Models (ELM) Research
OpenAI introduces Evolution through Large Models (ELM), a method that uses LLMs to improve genetic programming by approximating human-like code mutations to generate functional programs in novel domains.
Anthropic Softmax Linear Units (SoLU) Research
Anthropic introduces Softmax Linear Units (SoLU), an architectural change to MLP activation functions that increases the fraction of interpretable neurons without sacrificing model performance.
Intel and Hugging Face Partnership for Machine Learning Hardware Acceleration
Intel has joined the Hugging Face Hardware Partner Program to develop the Optimum Intel library, enabling streamlined hardware acceleration for training, fine-tuning, and inference of Transformer models on Intel platforms.
Hugging Face Director of Machine Learning Insights: Finance Edition
Experts from U.S. Bank, RBC, and Moody's Analytics discuss how machine learning is transforming finance through fraud detection and personalized products while navigating challenges like legacy systems and regulatory requirements.
AI-written critiques help humans notice flaws
OpenAI research demonstrates that AI-generated critiques help human evaluators identify significantly more flaws in summaries, suggesting a scalable path for human supervision of complex AI tasks.
OpenAI Techniques for Training Large Neural Networks
OpenAI outlines the primary parallelism strategies—data, pipeline, tensor, and expert parallelism—and memory-saving techniques used to orchestrate GPU clusters for training massive AI models.
Deep Q-Learning with Space Invaders
Hugging Face explains Deep Q-Learning (DQN), a reinforcement learning method that uses neural networks to approximate Q-values for high-dimensional state spaces like Atari games.
Annotated Diffusion Model – detailed walkthrough of DDPM implementation
Hugging Face released an annotated, step‑by‑step PyTorch implementation of Denoising Diffusion Probabilistic Models (DDPM), showing how to build, train, and sample from a diffusion model for image generation.
OpenAI, Cohere, and AI21 Labs Best Practices for Deploying Language Models
OpenAI, Cohere, and AI21 Labs have released a joint set of preliminary best practices for the responsible development and deployment of large language models to mitigate risks and prevent misuse.
OpenAI Research: Teaching Models to Express Their Uncertainty in Words
OpenAI researchers demonstrated that GPT-3 can be trained to express calibrated uncertainty about its answers in natural language without relying on model logits.
Graphcore and Hugging Face Expand IPU-Ready Transformer Models
Graphcore and Hugging Face have expanded the Hugging Face Optimum library to include 10 optimized transformer models across NLP, speech, and computer vision for Graphcore IPUs.
Hugging Face Hub Community Update: Pull Requests and Discussions
Hugging Face has introduced pull requests and discussions to the Hub, enabling community-wide collaboration on models, datasets, and Spaces without the need for repository forks.
OpenAI Codex Applications and Integration
OpenAI Codex is a natural language-to-code system based on GPT-3 that powers over 70 applications, including GitHub Copilot, to increase developer productivity by converting English instructions into code.
TAPEX: Efficient Table Pre-training via Neural SQL Execution
TAPEX is a table pre-training approach that uses synthetic SQL execution pairs to achieve state-of-the-art results on table question answering and fact verification with significantly higher efficiency than previous methods.
Anthropic Research: Scaling Laws and Interpretability of Learning from Repeated Data
Anthropic researchers found that repeating a small fraction of training data can severely degrade LLM performance by consuming model capacity for memorization and damaging generalization structures like induction heads.
Hugging Face Introduction to Q-Learning
Hugging Face provides a technical guide to Q-Learning, an off-policy value-based reinforcement learning method that uses a Q-table to determine optimal state-action pairs.
Hugging Face Multimodal Project Ethical Charter
Hugging Face has established an ethical charter for its multimodal learning project to integrate ethical principles, transparency, and fairness directly into the machine learning research lifecycle.
Sempre Health Case Study: Accelerating NLP Pipelines with Hugging Face Expert Acceleration Program
Sempre Health utilized the Hugging Face Expert Acceleration Program to deploy an NLP pipeline that automatically handles over 20% of inbound patient messages, significantly improving business scalability.
OpenAI DALL·E 2 Research Preview Update
OpenAI is expanding access to DALL·E 2 to waitlisted users at a rate of 1,000 people per week following a limited research preview that generated over 3 million images.
Hugging Face Deep RL Course Unit 2 Part 1: Introduction to Q-Learning Concepts
Hugging Face released the first part of its Deep Reinforcement Learning course, explaining value‑based methods, the Bellman equation, and Monte‑Carlo vs. Temporal‑Difference learning as groundwork for Q‑Learning.
Hugging Face Fellowship Program Announcement
Hugging Face has launched the Fellowship Program to support and empower key contributors to the open-source machine learning ecosystem through resources, recognition, and tailored support.
Machine Learning Experts Interview with Sasha Luccioni – Highlights on Ethical AI and Climate Impact
Hugging Face interviewed research scientist Sasha Luccioni, who discussed her work on ethical AI, carbon‑footprint measurement, and applying machine learning to climate and societal challenges.
Gradio 3.0 Release Notes
Hugging Face has released Gradio 3.0, featuring a complete frontend redesign and the introduction of Gradio Blocks for building complex, custom ML web apps in Python.
Hugging Face Student Ambassador Program 2022
Hugging Face has launched its first Student Ambassador Program to help achieve its goal of teaching machine learning to 5 million people by 2023.