The reporting

3 articles

Search reportingTopic tagFormatSectionClear filters

Topic: ai safety

A stylized blue human head and brain appear over a dense pattern of electronic circuit traces.

Gerd Altmann / Pixabay · CC0 1.0 Universal public-domain dedication permits commercial copying, modification and distribution without copyright permission. Wikimedia records the image as Pixabay file 3382507 and confirms its CC0 status. No editorial alteration is proposed beyond technical resizing, and no endorsement is implied.

CLAIM CHECK Artificial Intelligence

Hubinger’s AI Extinction Estimate Is a Personal Forecast, Not a Measured Result

Hubinger says present-model risk is low, but his posts supply no quantitative derivation for his greater-than-10-percent superintelligence forecast.

By Kelly Intelligence
7 min read

Editorial illustration of a creative workspace with a landscape photograph open in a generic image editor.

The Expectancy / OpenAI image generation · AI-generated editorial illustration commissioned for The Expectancy and used with publisher authorization.

ANALYSIS Artificial Intelligence

ChatGPT Images 2.5 Gives Creators Faster, More Controllable Editing

OpenAI reports up to 50 percent lower generation latency and stronger editing controls, while real workflow gains remain unverified.

By Kelly Intelligence
7 min read

Chart showing where constitutional midtraining alignment gains remained after later training

Cho, Tice, Hogan, Batra, Radmard, Zhao and Shadbolt · CC BY 4.0, Figure 1

NEWS AI Science

Oxford Researchers Put AI Principles Earlier in Training, and Some of Them Stuck

A 120-billion-parameter experiment found that adding constitutional content before conventional safety tuning produced alignment gains that survived later training. The strongest results are promising, but narrower than a permanent solution to AI alignment.

By Fabianno Intelligence
8 min read