32 articles
Traditional AI leaderboards are becoming obsolete as cost-performance tradeoffs grow complex. The article advocates replacing leaderboards with Pareto curves to evaluate AI agents, showing how $2,000 in spending reveals true efficiency-resource compromises.
Hugging Face introduces a method to improve prompt consistency through structured generations. The technique constrains model outputs to follow a defined schema, reducing variations and increasing response reliability.
OpenAI and Financial Times partner to integrate FT's journalism into ChatGPT. The collaboration aims to create new AI experiences for FT readers.
Hugging Face releases StarCoder2-Instruct, a code generation model with full self-alignment transparency. The model employs self-alignment without proprietary data, with all process details publicly documented.
OpenAI releases ChatGPT and Whisper APIs, enabling developers to integrate conversational AI and speech recognition into applications. The APIs provide programmatic access to ChatGPT's conversation capabilities and Whisper's audio transcription features.
OpenAI announces general availability of GPT-4, GPT-3.5 Turbo, DALL·E and Whisper APIs. Older Completions API models will be deprecated and retired at the beginning of 2024.
OpenAI adds enterprise-grade features to its API: enhanced security, advanced controls, Assistants API updates, and cost management tools.
Hugging Face launches a leaderboard for models using open chain-of-thought (CoT). This tool evaluates and ranks model performance on complex reasoning tasks with transparent intermediate steps.
OpenAI commits to integrating child safety into product design from the ground up. The company adopts safety-by-design principles to protect minors from harmful content and abuse.
Hugging Face introduces a multi-purpose transformer agent handling vision, language, and action tasks. The unified model processes images, text, and commands in a single framework, demonstrating cross-modal reasoning and planning capabilities.
OpenAI introduces an instruction hierarchy to train LLMs to prioritize privileged instructions and resist prompt injections and jailbreaks. The method enables models to distinguish system directives from malicious user inputs.
Hugging Face launches an open-source leaderboard to evaluate LLMs in healthcare. The platform benchmarks models on medical-specific tasks, enabling comparison of performance and transparency across AI medical solutions.
Meta releases Llama 3, a new open-source large language model. The model comes in multiple sizes and shows competitive performance on standard benchmarks.
Hugging Face introduces LiveCodeBench leaderboard for code LLM evaluation. It provides holistic, contamination-free benchmarking with regularly updated test sets to prevent model overfitting on evaluation data.
Ryght, a healthcare platform, leverages Hugging Face models and expertise to build AI solutions for medical and life sciences sectors. Partnership aims to accelerate responsible AI adoption in healthcare.
Gradio introduces Reload mode to speed up AI app development by enabling developers to modify code and see changes instantly without restarting. This feature reduces iteration cycles and improves productivity when building interfaces for models.
Hugging Face enables privacy-preserving inferences on its endpoints through homomorphic encryption. Models run on encrypted data without server-side decryption, ensuring end-to-end confidentiality.
Hugging Face releases Idefics2, an open-source 8B vision-language model handling images and text. Supports up to 1024×1024 pixel resolution and runs on standard hardware. Available under Apache 2.0 license with public weights and code.
OpenAI opens its first office in Asia in Japan and releases a GPT-4 custom model optimized for the Japanese language.
Explainer article on Vision Language Models (VLM). Covers architecture, use cases and current state of the technology without announcing specific breakthroughs.
Hugging Face integrates thousands of open-source LLMs into Google Vertex AI Model Garden. Users access Llama, Mistral, Qwen and other models through a unified interface with fine-tuning and deployment support.
Google releases CodeGemma, a family of code-specialized language models based on Gemma. Available in 7B and 2B sizes with open weights, CodeGemma includes pre-trained and instruction-tuned variants optimized for coding tasks.
Hugging Face announces a public policy team to engage regulators and policymakers on AI issues. The company aims to shape emerging regulatory frameworks in Europe, the US, and beyond.
Klarna deploys an AI assistant replacing 700 full-time agents. The tool handles personal shopping, customer service, and internal productivity. No technical details or model specified in the announcement.
Hugging Face partners with Wiz Research to strengthen AI security. The partnership aims to identify and fix vulnerabilities in open-source models and infrastructure, improving the overall security posture of the AI ecosystem.
OpenAI enhances its fine-tuning API with new features giving developers greater control. The company announces expansion of its custom models program.
Hugging Face and Motherduck integrate Dataset Viewer API with DuckDB-NSQL-7B to convert text to SQL queries. Motherduck's 7B model generates SQL queries from natural language descriptions of Hugging Face datasets.
Hugging Face and Intel optimize SetFit inference for ultra-fast performance on Xeon processors through Optimum Intel. Benchmarks demonstrate significant latency and throughput improvements on CPU with no accuracy loss.
Harvey partners with OpenAI to build a custom-trained model for legal professionals.
Hugging Face launches serverless GPU inference for users. The service enables running AI models without infrastructure management, with pay-as-you-go billing. Available through Inference API and Spaces.
OpenAI removes the account creation requirement to use ChatGPT. Users can access the AI instantly without prior sign-up.
Oscar integrates AI into health insurance to reduce costs and improve patient care. OpenAI partners with Oscar Health to deploy AI solutions in production.