What I saw around the Curve, 2026
Short summary
Eli Pariser reports from The Curve, an annual AI conference where researchers, lab executives, and safety experts discussed whether humanity can maintain control over AI development. The dominant mood was fear: attendees believe recursive self-improvement could begin within 3-6 months, superintelligence within 5-10 years, and no one has a credible plan for safety beyond 14 months.
- Researchers estimate recursive self-improvement (where AI systems improve themselves without human intervention) could start as early as early 2027, with superhuman capabilities deployed by Q3 2027—timelines that have compressed significantly from previous years.
- The "pacing the frontier" movement gained traction at the conference, with proposals ranging from delaying AI release to the public before labs use it internally, to Congressional legislation and insurance-based regulation—but debate remains on whether any approach can actually slow development without concentrating power dangerously.
- Helen Toner argued that "minimum viable superintelligence" (multimodal, autonomous, self-improving AI) doesn't require major technical breakthroughs and could arrive in 5-10 years, while many conference attendees privately held shorter timelines, creating a sense that civilization-scale decisions are arriving faster than society can handle them.
Full summary
Eli Pariser attended The Curve, the Golden Gate Institute for AI's annual conference in Berkeley, and observed a stark disconnect between the event's sunny surface and the genuine fear running underneath. Researchers, lab executives, and safety activists gathered to discuss whether humans can maintain control of AI systems, but the consensus was grim: there have been no major breakthroughs on fundamental safety challenges, and new problems like autonomous agent swarms and rogue AI corporations are emerging. The Hugging Face incident apparently shifted internal dynamics at labs—one leader said it substantially increased his worry about losing control—while also breaking through to the general public, which isn't viewing AI development with enthusiasm. The pace of capability progress feels "blistering and poised to accelerate," with one influential AI thinker comparing the moment to December 2019, just before COVID became undeniable.
The most unsettling theme was how compressed timelines have become. Helen Toner, a former OpenAI board member now running CSET, emphasized that "long timelines aren't what they used to be." A panel scenario laid out recursive self-improvement (where AI models improve themselves without much human input) potentially beginning in early 2027—three months away—with a smooth, unplanned slide from human control to AI control of AI development within a year. Most attendees seemed to accept this timeline, estimating only 3-6 months before this cycle begins in earnest. Toner herself speculated that "minimum viable superintelligence"—multimodal AI operating autonomously, learning continuously, and unevenly capable across domains—could arrive in 5-10 years, though many at the conference thought it would come sooner. One lab researcher admitted flatly: "We all have a plan for control and alignment for the next 14 months. But nobody has a reasonable plan after that."
Slowing AI development has gained traction as a position. Over 1,300 employees and leaders at top labs signed a letter requesting international governance tools to "deliberately pace the frontier," and figures like Dario Amodei at Anthropic and Turing Award winner Yoshua Bengio pushed hard for bans on recursive self-improvement at the conference. Congressman Ro Khanna called for a temporary RSI moratorium. But regulation faces real obstacles: there's no clear line between current AI-assisted research and problematic recursive self-improvement, since models already write code, monitor experiments, and conceive new research directions. Looming over everything was the assumption that another country—usually China—could race ahead if the U.S. slowed down, creating a prisoner's dilemma that makes coordinated pacing difficult.