✷ The archive · 11 labs · 3,061 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.
Qwen1.5-110B release notes / what's new
Qwen has released Qwen1.5-110B, the first model in the Qwen1.5 series to exceed 100 billion parameters, delivering competitive performance against Llama-3-70B and significant improvements over Qwen1.5-72B.
Moderna and OpenAI Partnership for AI-Driven Drug Development
Moderna has deployed ChatGPT Enterprise across its organization to accelerate mRNA medicine development and optimize business operations through custom GPTs.
OpenAI Introducing ChatGPT and Whisper APIs
OpenAI has released the GPT-3.5 Turbo and Whisper APIs, offering developers significant cost reductions and high-performance speech-to-text capabilities.
GPT-4 API General Availability and Completions API Deprecation
OpenAI has made the GPT-4 API generally available to all paying customers and announced a deprecation plan for older models in the Completions, Embeddings, and Edits APIs.
OpenAI Child Safety Commitment: Adopting Safety by Design Principles
OpenAI and other industry leaders have committed to implementing Safety by Design principles to mitigate generative AI risks to children, focusing on the prevention of child sexual abuse material (CSAM) and exploitation material (CSEM).
OpenAI Enterprise API Feature Updates April 2024
OpenAI has introduced new enterprise-grade security, administrative controls, Assistants API enhancements, and cost-management options to support scaling AI solutions for large organizations.
Hugging Face Open Chain of Thought Leaderboard
Hugging Face has introduced the Open Chain of Thought Leaderboard to measure the specific accuracy gain provided by chain-of-thought prompting across various LLMs on challenging reasoning tasks.
Anthropic Announces Child Safety Principles Initiative
Anthropic announced a partnership with Thorn and All Tech Is Human to adopt Safety by Design principles that aim to prevent generative AI from creating or spreading child sexual abuse material.
Anthropic Simple Probes Can Catch Sleeper Agents
Anthropic introduced linear “defection probes” that detect sleeper‑agent LLM behavior with >99% AUROC using only generic contrast pairs, showing a promising, low‑cost interpretability tool for AI safety.
Jack of All Trades (JAT) Multi-Purpose Transformer Agent
Hugging Face introduces Jack of All Trades (JAT), a single transformer-based agent capable of performing diverse sequential decision-making tasks across Atari, BabyAI, Meta-World, and MuJoCo environments.
OpenAI The Instruction Hierarchy: Prioritizing Privileged Instructions to Prevent Prompt Injection
OpenAI has introduced an instruction hierarchy that trains LLMs to prioritize system prompts over untrusted user input, significantly reducing vulnerability to prompt injections and jailbreaks.
The Open Medical-LLM Leaderboard: Benchmarking Large Language Models in Healthcare
Hugging Face has introduced the Open Medical-LLM Leaderboard, a standardized platform to evaluate and compare the performance of LLMs across diverse medical datasets to improve reliability and patient safety.
Llama 3 Refusal Rates and Censorship Comparison
Llama 3 exhibits significantly lower false refusal rates than Llama 2, reducing over-refusals on benign prompts while still providing technical assistance.
Meta Llama 3 Release Notes
Meta has released Llama 3, an open-access LLM family featuring 8B and 70B parameter models with improved tokenization and training on 15 trillion tokens.
Llama 3 Availability on Ollama
Ollama has released support for Llama 3, providing access to the 8B and 70B parameter models with significant improvements in training data, context length, and token efficiency over Llama 2.
Mixtral 8x22B Release Notes
Mistral AI has released Mixtral 8x22B, a sparse Mixture-of-Experts model with 141B total parameters and 39B active parameters, providing high performance and cost efficiency under an Apache 2.0 license.
CodeQwen1.5 Release Notes
Qwen has released CodeQwen1.5-7B, an open-source code LLM supporting 92 programming languages and 64K token context windows to enhance developer productivity.
Ryght Case Study: Building a Life Sciences Generative AI Platform with Hugging Face
Ryght has launched Ryght Preview, an enterprise-grade generative AI platform for healthcare and life sciences that leverages Hugging Face's Expert Support, TGI, and TEI to provide secure, flexible, and high-performance AI copilots.
Running Privacy-Preserving Inferences on Hugging Face Endpoints
Hugging Face and Zama have enabled the deployment of Fully Homomorphic Encryption (FHE) models via Hugging Face Endpoints, allowing users to perform machine learning inferences on encrypted data without decrypting it.
LiveCodeBench Leaderboard: Contamination-Free Evaluation for Code LLMs
Hugging Face has introduced the LiveCodeBench leaderboard, a new benchmark developed by researchers from UC Berkeley, MIT, and Cornell to evaluate LLM code generation and reasoning capabilities while preventing benchmark contamination using time-windowed problem sets.
Gradio Reload Mode for Faster AI App Development
Gradio's reload mode enables developers to apply source code changes to AI applications instantly without restarting the server, significantly reducing development latency.
Idefics2 8B Vision-Language Model Release – Architecture, Data, and Performance
Hugging Face released Idefics2, an 8B open‑source vision‑language model that outperforms other 8‑B models on VQA and OCR benchmarks and is ready for fine‑tuning via 🤗 Transformers.
OpenAI Japan Launch and GPT-4 Japanese Custom Model
OpenAI has opened its first Asian office in Tokyo and released a GPT-4 custom model optimized for the Japanese language that operates up to 3x faster than GPT-4 Turbo.
Vision Language Models Explained
Hugging Face provides a comprehensive guide to Vision Language Models (VLMs), detailing their architecture, open-source options, evaluation benchmarks, and new experimental support for fine-tuning via the TRL library.
Hugging Face and Google Cloud Vertex AI Model Garden Integration
Hugging Face has launched 'Deploy on Google Cloud,' enabling users to deploy thousands of open foundation models to Vertex AI or Google Kubernetes Engine (GKE) via the Hugging Face Hub or Vertex Model Garden.
CodeGemma Release Notes / What's New
Google has released CodeGemma, a family of open-access code-specialist LLMs based on Gemma, trained on 500 billion additional tokens of code, mathematics, and English language data.
Measuring the Persuasiveness of Language Models
Anthropic research demonstrates that AI persuasiveness scales with model size and capability, with Claude 3 Opus achieving a level of persuasiveness statistically comparable to human writers.
Hugging Face Public Policy Program Overview and Submitted Materials
Hugging Face announced a comprehensive public‑policy program that provides U.S., EU, and U.K. policymakers with detailed position papers, testimony, and comment letters, reflecting its cross‑functional commitment to responsible openness and shaping AI regulation.
Ollama Embedding Models Support
Ollama now supports embedding models, enabling the creation of local retrieval augmented generation (RAG) applications by converting text into vector embeddings.
Klarna AI Assistant Integration and Performance Results
Klarna has deployed an OpenAI-powered AI assistant that handles two-thirds of its customer service chats, performing the work of 700 full-time agents while improving resolution times and profit.
Hugging Face and Wiz Research Partnership for AI Security
Hugging Face has partnered with Wiz to integrate advanced vulnerability management and cloud security posture management to protect its platform and the broader AI/ML ecosystem.
Text2SQL with Hugging Face Dataset Viewer API and DuckDB-NSQL-7B
Hugging Face demonstrates how to use the DuckDB-NSQL-7B model and the Dataset Viewer API to convert natural language questions into SQL queries for analyzing over 120,000 open datasets.
OpenAI Fine-Tuning API Improvements and Custom Models Program Expansion
OpenAI has introduced new control features for its fine-tuning API and expanded its Custom Models program to include assisted fine-tuning and fully custom-trained models for domain-specific needs.
SetFit Inference Acceleration with 🤗 Optimum Intel on Xeon
Hugging Face demonstrates how to achieve up to 7.8x faster inference throughput for SetFit models on Intel Xeon CPUs using post-training static quantization via the 🤗 Optimum Intel library.
Qwen1.5-32B release notes / what's new
Qwen has released Qwen1.5-32B and Qwen1.5-32B-Chat, models designed to balance high performance with lower memory and inference costs compared to the 72B version.
Hugging Face and Cloudflare Workers AI Integration
Hugging Face integrated Cloudflare Workers AI to provide serverless GPU inference for popular open models, allowing developers to deploy AI applications with a pay-per-request pricing model.
OpenAI and Harvey Partner to Develop Custom Case Law Model
OpenAI and Harvey have collaborated to create a custom-trained case law model that incorporates 10 billion tokens of U.S. case law to improve reasoning and reduce hallucinations in legal professional tasks.
Anthropic Many-Shot Jailbreaking Research
Anthropic has identified a 'many-shot jailbreaking' vulnerability where providing a large number of faux dialogue examples in a long context window can override an LLM's safety training.
ChatGPT Access Update: Instant Use Without Account Sign-up
OpenAI has enabled the ability to use ChatGPT instantly without requiring an account sign-up, expanding accessibility to AI for a global audience.
Oscar Health AI Implementation: Reducing Costs and Improving Care
Oscar Health uses OpenAI's API to automate clinical documentation and claims processing, reducing documentation time by 40% and escalation resolution time by 50%.
OpenAI Voice Engine: Synthetic Voice Capabilities and Safety Framework
OpenAI has previewed Voice Engine, a model capable of creating realistic, emotive synthetic voices from a single 15-second audio sample, while maintaining a cautious deployment strategy to mitigate misuse.
Qwen1.5-MoE-A2.7B Release Notes
Qwen introduces Qwen1.5-MoE-A2.7B, a Mixture-of-Experts model that matches the performance of 7B dense models while using only 2.7 billion activated parameters.
OpenAI Zelma: Making Education Data Accessible via GPT-4
Zelma is a GPT-4 powered research assistant designed to make U.S. standardized test data for grades 3-8 accessible to parents, teachers, and policymakers through plain-language queries.
OpenAI Sora First Impressions from Creative Professionals
OpenAI shared early feedback from visual artists and filmmakers who are using Sora to prototype ideas, create surreal visuals, and remove technical and budgetary constraints from the creative process.
Pollen-Vision: Unified Interface for Zero-Shot Vision Models in Robotics
Hugging Face and the Pollen Robotics team have released pollen-vision, an open-source library that integrates zero-shot vision models to enable robots to detect and localize unknown objects in 3D space.
Anthropic Third-Party Testing AI Policy Proposal
Anthropic proposes a third-party testing regime for frontier AI systems to validate safety, prevent national security risks, and avoid regulatory capture through a diverse ecosystem of auditors.
Hugging Face Transformers: A Beginner's Guide to Open-Source ML
Hugging Face provides a comprehensive introductory guide to using the Transformers library and Hub to deploy and run open-source machine learning models like Microsoft's Phi-2.
Embedding Quantization: Binary and Scalar Techniques for Faster, Cheaper Retrieval
Hugging Face announced binary and int8 embedding quantization, cutting memory by 32× or 4× and speeding up retrieval up to 45× while keeping 96%–99% of original performance.
JetBrains AI Assistant Integration with OpenAI API
JetBrains integrated OpenAI's API into its AI Assistant to automate mundane coding tasks, resulting in 77% of developers reporting increased productivity.
Hugging Face and Lighthouz AI Introduce Chatbot Guardrails Arena
Hugging Face and Lighthouz AI have launched the Chatbot Guardrails Arena, a community-driven stress-testing platform designed to evaluate the data privacy and security of LLMs and their guardrails.