Technology news and discussion on TalkFuse.
The AI Safety Community Is Increasingly Freaked Out. Should They Be?
Warnings from parts of the AI safety community have become more urgent, with some researchers moving their estimates of potentially severe AI risks from a decade away to the next few years. The article says this shift matters because, if those shorter timelines are right, policymakers and researchers may have little time to gather stronger evidence or act.
A recent METR report found that AI development at Anthropic, measured through performance on defined tests, appears to be progressing roughly twice as fast as in previous years, largely with help from AI models. Anthropic co-founder Sam McCandlish broadly agreed, but the company withheld key data, limiting independent scrutiny. The article also raises questions about METR’s independence because its research depends on company access and some staff have professional or social ties to AI firms; it stresses that extreme near-term scenarios remain outliers and are contested.
In short
- AI safety warnings have shifted towards shorter timelines.
- METR estimates Anthropic’s measured progress is roughly twice as fast.
- Withheld data and ties to AI firms complicate independent scrutiny.
AI Government Politics Technology