✷ The archive · 11 labs · 3,427 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.
Aya Expanse Release: Advancing Multilingual LLM Performance
Hugging Face and Cohere For AI have released Aya Expanse, a family of 8B and 32B open-weight models that set new state-of-the-art benchmarks for multilingual performance.
Simplifying, stabilizing, and scaling continuous-time consistency models
OpenAI introduces sCM, a simplified continuous-time consistency model that achieves diffusion-level sample quality in just two sampling steps, providing a ~50x speedup in generation.
HUGS launch: zero‑configuration, hardware‑optimized inference for open LLMs
Hugging Face launched HUGS, a zero‑configuration, hardware‑optimized inference service for open‑source LLMs that runs on NVIDIA, AMD, and soon AWS Inferentia and Google TPUs, enabling enterprises to host models in‑house with an OpenAI‑compatible API.
CinePile 2.0 release: adversarial refinement boosts video QA dataset quality
CinePile 2.0 introduces an adversarial refinement pipeline that upgrades weak QA pairs into vision‑dependent questions, releasing both the improved dataset and the full code, and shows significant performance gains for commercial and open‑source video‑LLMs.
SynthID Text Integration in Transformers v4.46.0
Google DeepMind and Hugging Face have integrated SynthID Text into Transformers v4.46.0, providing a method to apply imperceptible watermarks to AI-generated text for detection via trained classifiers.
OpenAI and Microsoft Partner with Lenfest Institute for AI Collaborative and Fellowship Program
OpenAI and Microsoft have partnered with the Lenfest Institute for Journalism to provide $10 million in funding and credits to help local newsrooms implement AI for business sustainability and innovation.
Deploying Speech-to-Speech on Hugging Face Inference Endpoints
Hugging Face provides a guide for deploying its Speech-to-Speech (S2S) pipeline using custom Docker images on Inference Endpoints to handle high computational demands and reduce latency.
Outlines-core 0.1.0 release notes / what's new
Hugging Face and dottxt have released outlines-core 0.1.0, a Rust port of the Outlines core algorithms for structured generation that improves index compilation speed and portability.
Stable Diffusion 3.5 Large Integration with Diffusers
Hugging Face has integrated Stable Diffusion 3.5 Large, an 8B parameter model available in standard and timestep-distilled versions, into the Diffusers library.
Hugging Face partners with Protect AI to add Guardian scanner for model security
Hugging Face partnered with Protect AI to embed the Guardian scanner into the Hub, automatically detecting dangerous model serialization exploits and improving security for the entire ML community.
Transformers.js v3 release adds WebGPU acceleration, expanded model support, and server‑side JavaScript compatibility
Transformers.js v3 adds WebGPU acceleration, new quantization formats, support for 120 model architectures, and Node.js/Deno/Bun compatibility, enabling fast, on‑device inference in browsers and JavaScript runtimes.
Anthropic Claude 3.5 Sonnet and Claude 3.5 Haiku Release
Anthropic has released an upgraded Claude 3.5 Sonnet and the new Claude 3.5 Haiku, alongside a public beta for a groundbreaking 'computer use' capability that allows the model to interact with standard computer interfaces.
Llama 3.2 in Keras
Llama 3.2 is fully supported in Keras via keras-hub, allowing users to load Hugging Face checkpoints and run models across JAX, PyTorch, or TensorFlow backends.
IBM Granite 3.0 Models on Ollama
IBM Granite 3.0 models, including dense and Mixture of Experts (MoE) variants, are now available on Ollama under the Apache 2.0 license.
Anthropic Sabotage Evaluations for Frontier Models
Anthropic has introduced a new framework of sabotage evaluations to detect if AI models can mislead users, insert hidden bugs, hide capabilities, or undermine oversight systems.
OpenAI o1 Reasoning Models for Complex Problem Solving
OpenAI has introduced o1 reasoning models designed to solve complex problems across coding, strategy, and research domains.
Mistral AI releases Ministral 3B and 8B edge models
Mistral AI has introduced Ministral 3B and 8B, two state-of-the-art edge models designed for on-device computing, low-latency agentic workflows, and privacy-first local inference.
Hugging Face Transformers Gradient Accumulation Fix
Hugging Face has updated the Transformers Trainer to ensure gradient accumulation is mathematically equivalent to full batch training by correcting how losses are averaged across batches.
Anthropic Introduces Feature-Based Classifiers Using Dictionary Learning
Anthropic's interpretability team announced preliminary experiments using dictionary learning features as classifiers, offering a fast, interpretable alternative to fine-tuned language models.
OpenAI Evaluating fairness in ChatGPT study summary
OpenAI’s study finds that name‑based harmful stereotypes appear in less than 0.1% of ChatGPT responses and that overall answer quality is consistent across gender and racial name cues.
Anthropic Responsible Scaling Policy Update
Anthropic has updated its Responsible Scaling Policy (RSP) to introduce a more flexible risk governance framework with specific capability thresholds and proportional AI Safety Level (ASL) standards.
OpenAI MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering
OpenAI introduces MLE-bench, a benchmark using 75 Kaggle competitions to measure the machine learning engineering capabilities of AI agents, with o1-preview achieving bronze medal levels in 16.9% of tasks.
Gradio 5 Security Review
Hugging Face conducted a comprehensive security audit of Gradio 5 with Trail of Bits, fixing all identified vulnerabilities to ensure machine learning applications are safe by default.
AMD EPYC Turin CPU delivers 2× LLM inference throughput over Genoa
AMD’s 5th‑gen EPYC Turin CPU delivers roughly double the LLM inference throughput of Genoa, enabling lower latency and higher throughput for Hugging Face workloads.
Scaling AI Data Processing with Hugging Face and Dask
Hugging Face and Dask enable the scaling of AI-based data processing from small local samples to hundreds of millions of rows using distributed computing and multi-GPU parallel inference.
Gradio 5 Release Notes
Hugging Face has released Gradio 5, a production-ready framework for building performant, scalable, and secure machine learning web applications using Python.
OpenAI and Hearst Content Partnership
OpenAI has partnered with Hearst to integrate content from over 20 magazine brands and 40+ newspapers into its AI products, including ChatGPT, to provide users with more reliable and cited journalism.
Hugging Face Transformers 4.45.0 Dynamic Speculative Decoding
Hugging Face and Intel Labs introduced dynamic speculative decoding in Transformers 4.45.0, accelerating text generation by up to 2.7x by dynamically adjusting the number of draft tokens based on model confidence.
Anthropic U.S. Elections Readiness
Anthropic has implemented a multi-layered safety framework including usage policy updates, red-teaming, and redirects to authoritative voting sources to mitigate AI misuse during the 2024 U.S. elections.
Improving Parquet Deduplication on Hugging Face Hub
Hugging Face is optimizing its storage architecture to improve Parquet file deduplication, proposing content-defined row groups to reduce storage overhead during dataset updates.
Open FinLLM Leaderboard launch – comprehensive zero‑shot benchmark for financial language models
Hugging Face launched the Open FinLLM Leaderboard, a zero‑shot benchmark covering 40 finance‑specific tasks across seven categories to evaluate LLM readiness for real‑world financial applications.
OpenAI Canvas beta launch: collaborative writing and coding interface for ChatGPT
OpenAI introduced Canvas, a beta interface that lets ChatGPT collaborate on writing and coding projects with inline editing, version control, and specialized shortcuts, initially rolling out to Plus and Team users.
OpenAI Establishes $4 Billion Credit Facility for Financial Flexibility
OpenAI has established a $4 billion revolving credit facility with a consortium of global banks to increase liquidity and support the scaling of AI research and infrastructure.
Chinese AI Global Expansion Analysis
Chinese AI companies are accelerating international expansion due to domestic market saturation, intense price wars, and regulatory pressures, targeting Southeast Asia, the Middle East, and Western consumer markets.
OpenAI Funding Announcement October 2024
OpenAI has raised $6.6 billion in new funding at a $157 billion post-money valuation to accelerate frontier AI research and increase compute capacity.
OpenAI Realtime API Release
OpenAI has launched the Realtime API in public beta, enabling developers to build low-latency, multimodal speech-to-speech experiences using GPT-4o.
GPT-4o Vision Fine-Tuning API Release
OpenAI has introduced vision fine-tuning for GPT-4o, allowing developers to customize the model with image-text datasets to improve specialized visual understanding and object detection.
OpenAI Prompt Caching API Release
OpenAI has introduced Prompt Caching for GPT-4o, GPT-4o mini, o1-preview, and o1-mini, offering a 50% discount and reduced latency for reused input tokens.
OpenAI Model Distillation API Integration
OpenAI has introduced an integrated Model Distillation suite to allow developers to use outputs from frontier models like o1-preview and GPT-4o to fine-tune and improve the performance of smaller models like GPT-4o mini.
OpenAI and Altera: Creating Collaborative Digital Humans with GPT-4o
Altera has developed autonomous AI agents, termed digital humans, that can collaborate with people in environments like Minecraft using a brain-inspired architecture powered by GPT-4o.
BenCzechMark: A Comprehensive Evaluation Suite for Czech LLMs
Hugging Face and academic partners have released BenCzechMark, the first comprehensive evaluation suite for Czech language models, featuring 50 tasks across 9 categories and a novel duel-based scoring mechanism.
OpenAI Disrupts STORM-0817 Iran-Linked Malware Activity
OpenAI disabled accounts used by the Iran-based threat actor STORM-0817 to develop Android malware, scrape Instagram profiles, and perform reconnaissance on Pakistani cybersecurity professionals.
OpenAI Disrupts SweetSpecter China-Linked Cyber Activity
OpenAI identified and banned accounts linked to the China-based adversary SweetSpecter, who attempted to use ChatGPT for offensive cyber operations and targeted OpenAI employees with spear phishing attacks.
OpenAI Investigation: Fake Russian Troll Error Message Hoax
OpenAI has debunked a viral post claiming to expose a Russian troll account's GPT-4o error message, revealing the incident was a manually created hoax likely originating in the United States.
OpenAI Disrupts CyberAv3ngers Iran-linked Cyber Research Activity
OpenAI has banned accounts linked to the Iran-affiliated threat actor CyberAv3ngers, who used LLMs for reconnaissance, code debugging, and vulnerability research targeting industrial control systems.
OpenAI Disrupts Operation Stop News Russian Influence Activity
OpenAI banned a cluster of ChatGPT accounts used by a Russia-origin influence operation, dubbed Stop News, which used AI-generated text and images to mimic news outlets and establish deceptive partnerships.
OpenAI Disrupts Operation A2Z Multilingual Influence Activity
OpenAI banned a cluster of accounts using its API to run a multilingual influence operation, dubbed Operation A2Z, which leveraged AI to manage fake personas and generate political content across X and Facebook.
OpenAI Disrupts Bet Bot Gambling Spam Network
OpenAI banned a set of accounts using its API via an Israel-based startup to run a gambling spam network on X, utilizing AI-generated personas to lure users to gambling sites.
OpenAI Disrupts Operation STORM-2035 Iranian Influence Activity
OpenAI banned ChatGPT accounts used by an Iranian-origin actor, known as Storm-2035, to generate deceptive long-form articles and social media comments targeting the U.S. election and other global political issues.
OpenAI Disrupts Rwandan Election Political Commenting Network
OpenAI banned a network of ChatGPT accounts in Rwanda used to generate partisan election content and attempt to manipulate X trends through high-volume comment spamming.