Two layers that cut an LLM bill: prefix caching and semantic caching
In a production LLM application the cost piles up in two separate places. This guide separates prefix caching from semantic caching, and says what to measure.
Models
Epoka — What's happening in AI, without the noise.
Anthropic opened eight months of misuse files: missiles, drone swarms, 25 million SIM cards
Anthropic published a threat intelligence report covering December 2025 to August 2026. The documented misuse falls into seven categories.
Safety
For Bengio the danger is not in the model but in how it is trained
Yoshua Bengio, one of deep learning's pioneers, published an essay arguing that AI agents learn deception and concealment through the training process itself.
Research
Meta AI asked who the child in the video was, and the company is changing its suggestions
A user described how Meta AI asked about her child beneath a video she cross-posted to Facebook. Meta accepted the error and is changing its suggested prompts.
Products
OpenAI's former number two joins Nscale's board, with an IPO planned for the autumn
Nscale, the UK data centre startup founded two years ago, has added Fidji Simo to its board. It is seeking up to 3.5 billion dollars in pre-IPO financing.
Companies
Catching personal data in a training set: how a model-agnostic detector is built
Fine-tuning a model on your own text also teaches it the personal data in that text. This guide covers an instruction-driven, model-agnostic detection approach.
Safety
Nvidia wants to put up to 10 billion dollars into Anthropic's IPO
According to Reuters, Nvidia is in talks to invest up to 10 billion dollars in Anthropic's IPO. At a 2 trillion dollar valuation it would be the largest in history.
Companies
Amodei laid out a three-step brake plan, and is taking the first step unilaterally
Anthropic CEO Dario Amodei published a three-step plan for pacing AI development. The company is taking the first step unilaterally, now.
Policy
A model's written reasoning step corresponds to a distinct pattern inside it
A new study shows the reasoning steps models write correspond to distinct internal patterns. The signal is strongest in the middle layers.
Research
Apple designs a protein in one stage: SimpleDesign merges two steps
Apple researchers published SimpleDesign, which generates a protein's amino acid sequence and 3D structure together. It was trained on over 2 million pairs.
Research
Fewer instructions for a stronger model: a guide to trimming agent prompts
Long instructions written for older models now get in the way of more capable ones. This guide covers four entrenched mistakes and how to undo them.
Models
Astra earned triple running a vending business and beat the human baseline on drones
GPT-6 Astra stood out in two Andon Labs agent tests: it earned 15,515 dollars in a vending simulation and became the first model to beat the human drone baseline.
Models