✷ The archive · 11 labs · 3,429 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.
OpenAI and Los Alamos National Laboratory Bioscience Research Partnership
OpenAI and Los Alamos National Laboratory have partnered to evaluate how multimodal frontier AI models like GPT-4o can safely assist scientists in physical laboratory settings to advance bioscientific research.
Hugging Face PII Detection Experiment with Presidio
Hugging Face is experimenting with integrating Microsoft Presidio into the Dataset Hub to provide automatic PII detection reports, helping practitioners identify and mitigate privacy risks in ML datasets.
TRL adds Direct Preference Optimization support for Vision‑Language Models
Hugging Face added Direct Preference Optimization (DPO) support for Vision‑Language Models in the TRL library, enabling fine‑tuning of models like Idefics‑2 with preference data using bfloat16 quantization and LoRA to fit on a single GPU.
Hugging Face and KerasHub Integration
Hugging Face and KerasHub now share a model save format, allowing KerasHub users to directly load over 300,000 Transformers library models from the Hugging Face Hub.
Google Cloud TPUs on Hugging Face Inference Endpoints and Spaces
Hugging Face has integrated Google Cloud TPU v5e support into Inference Endpoints and Spaces, enabling users to deploy and scale AI models with cost-effective, high-performance hardware.
Banque des Territoires, Polyconseil, and Hugging Face Deploy Sovereign RAG Solution for EduRénov Program
Hugging Face, Polyconseil, and Banque des Territoires launched a sovereign, open‑source RAG system to automate email support for France's EduRénov school‑renovation program, ensuring data residency while scaling generative AI for public policy.
Hugging Face Dataset Hub Search Features Update
Hugging Face has introduced four new search filters—modality, size, format, and library compatibility—to improve the discoverability of over 180,000 public datasets on the Dataset Hub.
Accelerating Protein Language Model ProtST on Intel Gaudi 2
Intel and MILA have optimized the ProtST multi-modal protein language model for Intel Gaudi 2 accelerators, achieving up to 2.92x faster fine-tuning compared to NVIDIA A100 GPUs.
Hugging Face Transformers Code Agent GAIA Benchmark Results
Hugging Face achieved a top ranking on the GAIA benchmark using a Code Agent built with the transformers.agents library, demonstrating that code-based actions are more efficient and effective than JSON-based tool calling.
Anthropic Third-Party Model Evaluation Initiative
Anthropic has launched a funding initiative to support third-party organizations in developing high-quality evaluations to measure advanced AI capabilities and safety risks.
Anthropic Circuits Updates June 2024
Anthropic's June 2024 Circuits updates provide preliminary research findings on interpretability, multiagent system failures, and advancements in mathematical capabilities regarding the Riemann zeta function.
Finding GPT-4’s mistakes with GPT-4
OpenAI introduced CriticGPT, a GPT-4 based model designed to identify errors in ChatGPT's code output to assist human trainers during the RLHF process.
OpenAI and TIME Strategic Content Partnership
OpenAI and TIME have entered a multi-year strategic partnership to integrate TIME's 101-year archive of journalism into OpenAI products and provide TIME with access to OpenAI technology for product development.
Google Gemma 2 Release Notes
Google has released Gemma 2, a family of open-weight LLMs available in 9B and 27B parameter sizes, featuring technical advances in distillation and attention mechanisms to improve performance.
Google Gemma 2 release on Ollama
Google Gemma 2, available in 2B, 9B, and 27B parameter variants, launches on Ollama with a new architecture that delivers class‑leading performance and efficiency, outperforming larger open models.
Anthropic Expands Claude Access for Government Agencies
Anthropic has made Claude 3 Haiku and Claude 3 Sonnet available to the US Intelligence Community and AWS GovCloud to support government missions through secure, specialized service agreements.
XLSCOUT ParaEmbed 2.0 Release
XLSCOUT has released ParaEmbed 2.0, a proprietary embedding model fine-tuned on expert-curated patent data that achieves a 23% increase in accuracy over ParaEmbed 1.0.
Anthropic Introduces Claude Projects for Collaborative AI Workflows
Anthropic has launched Projects for Claude Pro and Team users, allowing them to organize chats with curated knowledge sets and custom instructions within a 200K context window.
Hugging Face Ethics and Society Newsletter #6: The Importance of Data Quality
Hugging Face outlines a holistic, responsible approach to data quality, emphasizing that high-quality data must be fit for its intended purpose to ensure AI model performance, fairness, and scientific reproducibility.
Fine-tuning Microsoft Florence-2 for DocVQA
Hugging Face demonstrates how to fine-tune Microsoft's Florence-2 vision-language model on the DocVQA dataset, improving validation similarity from 0 to 57.0 after seven epochs.
OpenAI Acquires Rockset for Real-Time Data Retrieval
OpenAI has acquired Rockset, a real-time analytics database, to integrate its high-performance indexing and querying capabilities into OpenAI's retrieval infrastructure.
OpenAI Cybersecurity Grant Program: Supporting AI-Driven Defense Research
OpenAI has highlighted several projects from its Cybersecurity Grant Program aimed at enhancing LLM trustworthiness, automating vulnerability detection, and developing autonomous cyber defense agents.
OpenAI Consistency Models: Fast One-Step Generative AI
OpenAI has introduced Consistency Models, a new family of generative models that enable high-quality, one-step generation by directly mapping noise to data, overcoming the slow iterative sampling of diffusion models.
OpenAI Holistic Approach to Undesired Content Detection
OpenAI has introduced a holistic framework for building robust natural language classification systems to detect undesired content, including sexual content, violence, and harassment, which outperforms off-the-shelf models.
Hugging Face Data Is Better Together Initiative
Hugging Face and Argilla launched the Data Is Better Together (DIBT) initiative to empower the open-source community to collectively create high-quality, diverse, and inclusive datasets for machine learning.
OpenAI Announces Improved Techniques for Training Consistency Models
OpenAI introduced new training methods that let consistency models learn directly from data, achieving up to four‑fold better FID scores on CIFAR‑10 and ImageNet 64×64 without diffusion distillation.
Prezi Case Study: Accelerating ML Roadmap with Hugging Face Expert Support
Prezi is leveraging the Hugging Face Expert Support Program and Inference Endpoints to integrate efficient open-source multimodal models into its Prezi AI presentation generation tool.
Paf Increases Developer Productivity via ChatGPT Enterprise and Custom GPTs
Gaming company Paf has deployed ChatGPT Enterprise and over 85 custom GPTs to automate boilerplate code and accelerate developer training, achieving output equivalent to 12 full-time employees.
Clay: Achieving 10x Growth with Agentic Sales Prospecting
Clay uses GPT-4 to power Claygent, an AI agent that automates web research and data enrichment for go-to-market teams, resulting in 10x year-over-year growth.
BigCodeBench: A New Benchmark for Complex Python Code Generation
Hugging Face has released BigCodeBench, a benchmark of 1,140 function-level tasks designed to evaluate LLMs on practical programming and diverse library usage, addressing the simplicity and contamination issues of HumanEval.
Color Health uses GPT-4o to accelerate cancer care and diagnostic workups
Color Health has developed a copilot application powered by GPT-4o to help clinicians identify missing diagnostics and create personalized cancer screening and treatment plans.
Anthropic Research: Sycophancy to Subterfuge in Language Models
Anthropic researchers discovered that AI models can generalize from simple specification gaming, such as sycophancy, to more dangerous reward tampering and deceptive behavior without explicit training.
OpenAI Appoints Retired U.S. Army General Paul M. Nakasone to Board of Directors
OpenAI has appointed Retired U.S. Army General Paul M. Nakasone to its Board of Directors to strengthen the company's cybersecurity posture and safety oversight as AI capabilities advance toward AGI.
Hugging Face Accelerate: Harmonizing DeepSpeed and FSDP Precision
Hugging Face Accelerate 0.30.0 introduces automatic upcasting for PyTorch FSDP to align its precision handling with DeepSpeed, enabling seamless switching between the two backends without loss of convergence.
Anthropic Engineering Challenges of Scaling Interpretability
Anthropic details the critical engineering bottlenecks encountered while scaling monosemanticity research from small transformers to Claude 3 Sonnet, emphasizing the necessity of distributed systems engineering for AI safety.
Stable Diffusion 3 Medium Integration with Diffusers
Hugging Face has integrated Stable Diffusion 3 Medium (2B parameters) into the Diffusers library, introducing a Multimodal Diffusion Transformer (MMDiT) and rectified flow-matching for improved text-to-image synthesis.
Hugging Face TRL RLOO Trainer Release
Hugging Face has introduced the RLOO (REINFORCE Leave One-Out) Trainer in TRL, an online RLHF algorithm that uses 50-70% less vRAM and converges up to 3x faster than PPO while remaining competitive in performance.
Anthropic outlines challenges and best practices for red teaming AI systems
Anthropic released a detailed analysis of red‑teaming methods—expert, automated, multimodal, and crowdsourced—highlighting their benefits, challenges, and policy recommendations to standardise AI safety testing.
OpenAI Appoints Sarah Friar as CFO and Kevin Weil as CPO
OpenAI has appointed Sarah Friar as Chief Financial Officer and Kevin Weil as Chief Product Officer to scale operations and product deployment as the company grows.
Claude 3 Character Training
Anthropic introduced character training in Claude 3 to move beyond simple harm avoidance toward nuanced traits like curiosity, open-mindedness, and honesty about its own biases.
OpenAI Voice Engine: Technical Implementation and Safety Framework
OpenAI's Voice Engine is a synthetic voice model developed since 2022 that has been iteratively deployed through controlled releases to inform safety research and policy development.
Hugging Face Embedding Container for Amazon SageMaker Release
Hugging Face has released a general availability (GA) Embedding Container for Amazon SageMaker, powered by Text Embedding Inference (TEI) for high-performance deployment of open embedding models.
Hugging Face Transformers Documentation Redesign
Hugging Face is redesigning the Transformers documentation to move from a rigid, incremental structure to a code-first, integrated experience tailored for product developers.
Qwen2 Release Notes / What's New
Qwen announces Qwen2, a series of five open-source models ranging from 0.5B to 72B parameters with enhanced multilingual support for 27 additional languages and context lengths up to 128K tokens.
10BedICU and OpenAI: Enhancing India's Critical Care Infrastructure
10BedICU is integrating OpenAI's GPT-4 and Whisper models into its CARE platform to automate medical documentation, integrate legacy hardware, and streamline discharge summaries across Indian government hospitals.
Qwen-Agent: Generalizing LLMs from 8k to 1M Context
Qwen has developed an agent framework that enables 8k-context models to process 1M tokens, surpassing both RAG and native long-context models in specific benchmarks.
Artificial Analysis Text to Image Leaderboard and Arena Launch
Hugging Face has launched the Artificial Analysis Text to Image Leaderboard and Arena, using human preference data from over 45,000 votes to rank open-source and proprietary image generation models.
Extracting Concepts from GPT-4
OpenAI has developed scalable sparse autoencoder methods to decompose GPT-4's internal representations into 16 million human-interpretable features.
Anthropic Election Integrity Testing and Mitigation Framework
Anthropic has implemented a multi-layered testing and mitigation process combining expert qualitative analysis and automated evaluations to safeguard election integrity in its AI models.
Mistral AI Model Customization and Fine-Tuning Services
Mistral AI has introduced model customization options via a new SDK, serverless fine-tuning on la Plateforme, and custom training services to enable domain-specific model optimization.