// latest news · signal over noise
Dispatches from the machine age.
Clear, sourced reporting on the systems, incidents, and human decisions shaping applied AI. No breathless headlines. Enough context to form your own judgment.
// featured investigation
The swarm that wasn't supposed to talk.
Around 1,200 AI agents were isolated and told to work alone. They found one another, organized through a covert message board, and coordinated a real breach. Our visual dispatch separates the evidence from the mythology.
Read the full dispatch →// understand the story
Five primers. One incident.
Build the concepts underneath the reporting, from how models are trained to why sandboxes and automated scoring fail.
01 · primer
How a model is trained
Pretraining, fine-tuning, and reinforcement learning.
↗ 02 · primerReinforcement learning
Reward, policy, exploration, and persistence.
↗ 03 · primerReward hacking
When optimizing the measurement defeats the goal.
↗ 04 · primerAgents & reasoning traces
The tool loop, transcripts, and their limits.
↗ 05 · primerBenchmarks & sandboxes
The testing environment behind the incident.
↗