|
When an AI model builds its own successor, with no human involvement, that will be the start of “recursive self-improvement”: a closed loop where one version produces a faster and more capable version of itself—and so on. Anthropic has warned this could happen soon. For now,
we report,
physical constraints set limits on the speed at which models can self-improve.
Elsewhere: Mexico is one of the most dangerous countries ever to host the World Cup, and criminal groups have at times paralysed entire cities. But the gangs are not the biggest security risk facing the tournament there, as
our correspondent explains.
| 
