Inferize

The US military has moved to disable advertising trackers on official websites amid fears the technology could expose service members' locations and habits. The concern is that ad tracking data, gathered through embedded scripts, could be harvested by adversaries and used to profile troops or infer sensitive movements, a risk heightened by current tensions with Iran.

A project called Strata, shared on GitHub, claims it can run Qwen's 3.8 Flash Next 125-billion-parameter model on a single consumer RTX 4090 GPU at roughly 100 tokens per second. If verified, that would make a very large language model practical on high-end home hardware without a data center. The claim is drawing attention among developers interested in local AI inference, though independent confirmation of the speed figures has not been established.

Attention is turning to the economics of running large language models as a hosted service. Serving AI inference is costly, requiring expensive GPUs, heavy energy use and constant capacity planning, yet providers often price access aggressively to win users. The discussion explores why inference providers can operate at thin or negative margins, how GPU supply and demand shape pricing, and what this means for the sustainability of the booming AI services market.

A new write-up from Unblocked describes adaptive routing of large language model requests across multiple inference providers using an approach modelled on TCP congestion control. The system treats providers like network paths, backing off when a provider slows down and shifting traffic toward faster responses. The approach is drawing attention among developers interested in reliability and latency for LLM applications.

A new analysis tackles one of the startup world's thorniest economics problems: how to set subscription prices for AI coding agents whose compute costs can quickly exceed what customers pay. The piece weighs flat-rate versus usage-based models, noting that heavy users of agentic coding tools can burn through far more in inference costs than a typical monthly fee covers. Founders and investors are debating sustainable pricing as AI coding tools go mainstream.

Astronomers have detected radio waves from a planet outside our Solar System for the first time, an unprecedented observation. The signal suggests it may be possible to study the magnetic fields and environments of distant exoplanets directly, rather than inferring them from light passing through their atmospheres. Scientists caution the detection needs confirmation, but if verified it would open a new window on worlds orbiting other stars.

PCMag tested a privacy tool designed to show what personal information ChatGPT holds about a user, and found the results disturbingly accurate. The experiment highlights growing concern over how much data AI assistants can infer or retain about individuals, and is prompting readers to reconsider what they share with chatbots and how AI companies handle personal data.

A new essay argues that philosophy and theology together support an inference to design, framing the design argument as a serious philosophical position rather than a purely scientific claim. The piece is being circulated among readers interested in science-and-religion debates, where arguments for design remain a recurring point of contention.