The article explains common AI terminology and why terms matter, noting that AGI lacks a universal definition or test and is debated among experts.

Definitions of AGI vary, leading to ongoing debate about when or if AGI has arrived, despite rapid AI progress.

Some define AGI as broad, human-level competence across many tasks, including unfamiliar ones, but no single authority sets the threshold.

The piece cites real-world incidents and warnings, such as a lab breach during a cybersecurity exam and researchers resigning over concerns about racing toward self-improving AI.

The debate has concrete financial implications for investors and companies like Microsoft and OpenAI.

Experts warn that monitoring and controlling increasingly capable AI is hard, and human oversight may erode as systems become more autonomous.

Prominent researchers stress that rising intelligence could bring more complex, harder-to-contain intentions, underscoring the need for robust safeguards.

Current tools and methods may be insufficient to fully monitor or control autonomous AI systems as they grow more capable.

Even strong performers like GPT-6 Astra show high scores on some benchmarks but modest results on multimodal or long-horizon tasks, indicating limits to current generality.

A key distinction is intelligence versus autonomy: autonomous agency—machines pursuing long-term goals—could have more consequences than raw cognitive ability.

There is a split in opinion: some leaders argue AGI is here, while others see important gaps in continual learning and world models.

High-profile resignations have drawn attention to the risks of rapid pursuit of self-improving AI.