Artificial Intelligence / Analysis
Open weights do not answer every question about open AI
Access to model weights can enable experimentation and local deployment. Training transparency, licensing and operational control are separate issues.
THE TOPIC FILE
Artificial intelligence
From language models to autonomous workflows, understand how AI systems work, where they fail and which claims deserve scrutiny. Our coverage brings together model evaluation, deployment choices and the security implications of connecting AI to real information and tools.
Stories in this collection
Technology / Explainer
Why AI performance depends on moving data as well as doing arithmetic
Memory capacity, bandwidth, and compute throughput answer different questions. A useful performance comparison starts with the workload.
Artificial Intelligence / Explainer
A bigger context window is not a better memory
Long-context models can accept more information. Whether they use the right information remains a separate question.
Artificial Intelligence / Explainer
What RAG actually solves, and what it leaves behind
Retrieval-augmented generation can connect an AI model to reliable information. The difficult part is deciding which information deserves to be retrieved.
Artificial Intelligence / Explainer
Multimodal AI needs more than eyes
Reading a picture, a chart and a document requires different checks. A model that accepts all three has not necessarily understood them.
AI Security / Analysis
Shadow AI is an information problem before it is a tool problem
Employees adopt useful assistants faster than policies can name them. A workable response starts with the data and the task.
Software / Field guide
Review AI-generated code against the behavior the change is meant to preserve
A useful review connects requirements, implementation, and evidence. Plausible code and passing checks are only parts of that connection.
Artificial Intelligence / Analysis
What an AI benchmark leaves out
A leaderboard measures a test under particular conditions. It does not settle which system will work best inside your organization.
Artificial Intelligence / News analysis
Google's Search-agent plan changed the question from finding information to monitoring it
The May I/O announcements opened a useful debate about persistent search, notification quality, and the difference between a promise and a deployed service.
Technology / Explainer
On-device AI privacy starts with tracing the entire data path
Where a model runs matters, but a useful privacy assessment also follows inputs, outputs, retained records, and the services surrounding the model.
Companies / News analysis
Microsoft and OpenAI's revised agreement separated several kinds of dependence
The April amendment is a useful case study in why cloud access, intellectual-property rights, revenue sharing, and ownership should be read separately.
Artificial Intelligence / News analysis
Muse Spark made Meta's product choices part of the model story
A retrospective on the April announcement and the questions it raised about distribution, access, and the meaning of a useful assistant.