Every article we have published, newest first. Search titles and leads, or change the order.
ModelsNews
IBM Research released an open recipe that combines four Granite time-series models. You no longer have to guess which architecture suits your data: the ensemble returns median and quantile forecasts with no retraining.
· 2 min read
ToolsNews
Perplexity Computer can now run entirely on an Apple silicon Mac. The model, the harness and task execution all stay on the machine. Perplexity recommends at least 36 GB of memory.
· 2 min read
ResearchNews
UC Berkeley researchers train humanoid motion tracking through one differentiable world model. On a single GPU, they pass a baseline trained for 7 days on 128 GPUs in 3 hours.
· 2 min read
ResearchNews
Google DeepMind and Google Research released ForTy 2020 under an open license, a map that sorts every 10x10 meter patch of Earth into six forest and land-use types. An early yearly time series for 2017 to 2025 ships alongside it.
· 3 min read
ModelsNews
NVIDIA's Earth-2 open model family adds StormScope EU, trained on European data, which forecasts cloud conditions from Meteosat imagery at 10-minute intervals.
· 2 min read
ModelsNews
At GTC Berlin NVIDIA introduced StormScope EU, the European short-range forecaster in its Earth-2 weather model family. It predicts cloud conditions from Meteosat data every 10 minutes, under an open license, and runs on your own GPUs.
· 3 min read
BusinessNews
At GTC Berlin, NVIDIA showcased European government projects running open models on infrastructure they control. For developers, the two new synthetic datasets stand out: one million U.K. and German persona records each.
· 2 min read
BusinessNews
At GTC Berlin, European banks and payment companies showed how they run open weight models trained on their own data, on infrastructure they control. bunq says open models match leading closed models at 60% of the cost per answer.
· 2 min read
ToolsNews
OpenAI is updating its Usage Policies: from November 20, model routers and AI gateways that offer direct access to its models will need OpenAI's prior approval.
· 2 min read
BusinessNews
OpenAI's frontier models and Codex agentic workflows are coming to Google Cloud's Model Garden. General availability is planned in the coming weeks.
· 2 min read
BusinessNews
OpenAI engineers explained how ChatGPT's ad system works. At its core is a post-trained LLM that drives retrieval, ranking and the auction through tool calls.
· 2 min read
ToolsNews
Through OpenAI's Patch the Planet initiative, frontier AI found two critical remote code execution flaws in PHP. Fixes shipped on Wednesday, and OpenAI is proposing a five-stage framework for handling disclosures like this.
· 4 min read
ModelsNews
IBM updated three models in its Granite Speech family. The new granite-speech-5.0-2b is more accurate and noise-robust, Turbo CTC went multilingual, and 4.1 2B Plus is better at tracking who said what and when.
· 3 min read
ResearchNews
OpenAI, PerturbAI and Scripps Research ran thousands of parallel agents over a whole-brain mouse CRISPR atlas. The analysis took 32 minutes and surfaced new research leads.
· 2 min read
ModelsNews
NVIDIA released Kumo Relational, a model that predicts churn, demand or fraud directly on relational databases, with no task-specific training or feature engineering.
· 3 min read
ModelsNews
NVIDIA detailed Kumo Forecast, a time-series model that refines predictions with patterns retrieved from your own historical data and explains its forecasts. It ships under Apache 2.0.
· 2 min read
ModelsNews
NVIDIA and the University of Washington released an open model that generates protein sequence and all-atom structure together. You can try it in the browser, free until November 4.
· 2 min read
BusinessNews
Google Cloud will make Vera Rubin NVL72-based A5X instances generally available next month, with Europe to follow. It plans to scale to as many as 960,000 Rubin GPUs.
· 2 min read
ModelsNews
Mistral has offered Mistral Large 4 in public preview since October 6. The 1.05 trillion-parameter multimodal model handles a 1 million-token context. Open weights are promised by the end of October, with no license announced yet.
· 3 min read
ToolsNews
At GTC Berlin, NVIDIA released the OpenShell Reference Architecture, a full-stack blueprint for running agents inside contained, monitored boundaries. Policies are enforced in hardware by Sentry on BlueField DPUs.
· 2 min read
