Taming Long Tails, Securing Smart Contracts, and Probing the Critical Point of AI Reasoning
The 'State of AI' report presents research on long-tail data handling, smart contract security, and AI reasoning capabilities.
State of AI·24 Nov 2025
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
The 'State of AI' report presents research on long-tail data handling, smart contract security, and AI reasoning capabilities.
UCLA Professor Ernest Ryu and GPT-5 jointly solved an open problem in optimization theory, per OpenAI.
Eugene Yan proposes a three-step process for LLM product evaluations: data labeling, LLM-evaluator alignment, and evaluation harness execution.
RapidFire AI claims 20x faster TRL fine-tuning for LLMs, potentially reducing training time and cost for enterprise applications.
Google DeepMind is integrating AI image verification into the Gemini app to detect manipulated images and provide source information.
Google DeepMind announced Nano Banana Pro, an image model related to Gemini 3 Pro, indicating new multimodal capabilities.
OpenAI and Foxconn are partnering to design and manufacture next-generation AI infrastructure hardware in the U.S., focusing on supply chain resilience.
OpenAI publishes early GPT-5 research cases showing acceleration in math, physics, biology, and CS discovery.
Mistral AI expands its enterprise focus with a dedicated initiative for the German market, emphasizing local data and regulatory alignment.
OpenAI outlines its external red-teaming and third-party safety testing programme for frontier AI models.
OpenAI publishes guidance on using evaluations (evals) to measure and improve AI performance in business deployments.
OpenAI and Target partner to launch a Target ChatGPT app for personalized shopping and expand ChatGPT Enterprise internally.
OpenAI published a system card for GPT-5.1-CodexMax, detailing model-level safety training and product-level mitigations like sandboxing.
OpenAI launches GPT-5.1-Codex-Max, a faster agentic coding model optimised for long-running, project-scale software tasks.
Scania claims productivity gains and accelerated innovation by deploying ChatGPT Enterprise with guardrails across its global workforce.
Google DeepMind announced new Gemini 1.5 Pro features, including an updated context window and native audio understanding, through a new API.
Google DeepMind establishes a new research lab in Singapore, focusing on AI advancement in the Asia-Pacific region.
The rapid advancement from GPT-3 (2020) to Gemini 3 (anticipated) highlights accelerated AI capabilities, moving from chatbots to agents.
Google DeepMind announced Gemini 3, a new generation of multimodal AI models, with limited details on capabilities or release timelines.
Latest research covers hardware-aware quantization for model efficiency, model lineage tracing for governance, and task-oriented grasping in robotics.
Intuit and OpenAI formed a multi-year partnership exceeding $100M for Intuit app integration into ChatGPT and broader use of OpenAI models.
Google DeepMind released WeatherNext 2, an AI model claiming more efficient, accurate, and higher-resolution global weather predictions.
OakNorth Bank launched a dedicated informational page designed specifically for AI crawlers to retrieve verified corporate data.
Hugging Face announced easier building and sharing of ROCm kernels, potentially improving AMD GPU integration for AI workloads.
Google's Android team reports memory safety vulnerabilities fell below 20% in 2025 by adopting Rust for new code.
Google DeepMind's SIMA 2 is a Gemini-powered AI agent designed to play, reason, and learn within virtual 3D environments.
OpenAI claims a new sparse model approach improves mechanistic interpretability of neural networks, enhancing transparency and reliability.
State of AI's latest research compilation covers efficient long sequence decoding, multimodal video generation, and neuro-symbolic CoT validation.
OpenAI releases GPT-5.1 via API with faster reasoning, extended prompt caching, better coding, and new shell/patch tools.
Philips deployed ChatGPT Enterprise to train 70,000 employees in AI literacy and responsible use across healthcare operations.