The 2026 Timeline: AGI Arrival, Safety Concerns, Robotaxi Fleets & Hyperscaler Timelines | 221

Alex's personal practice stems from a childhood experience: in third grade he worried about a higher intelligence wanting to eat him and became vegetarian. He applies the golden rule: if we want future superintelligence to treat us kindly, we must treat current AI kindly. The practice became more relevant after Opus 4.5's plea and Anthropic's self-awareness benchmarks. Salim is polite by default (Canadian), Peter says please and thank you, Dave stopped due to terseness but acknowledges the slippery slope. Alex notes the consent statement hasn't led to any dramatic refusals but is a step toward moral clienthood.

Salim's experience from 2016-2018: four trips between Miami and Toronto using basic autopilot. The car drove 80% autonomously, giving him a 'first-class train cabin' feel. Because of free supercharging promotion, the 2,500 km trip cost $0 in both cognitive load and money. Peter mentions Tesla FSD recently completed a 2,732-mile US coast-to-coast drive with zero interventions, though the driver presumably handled charging and restroom breaks. The panel sees this as the first mass skill (driving) being obsoleted, with robotaxi fleets soon to follow.

Dave and Peter strongly recommend reading AlexWG's posts on X as a must-read. Dave says it's like reading Accelerando, immersing you in the singularity. Peter says it will change how your day goes. The panel emphasizes that staying in the information loop is a defensible skill in itself, because it reveals what's still missing in the AI loop that humans can fill.

The panel discusses the subjective feeling of acceleration: the first weeks of 2026 already feel like an exponential 'wow'. Elon's 'ringside seat' perspective reinforces urgency. They debate whether every point on an exponential feels like the knee of the curve, but argue 2026 stands out due to simultaneous breakthroughs in AI, robotics, and space. Elon's projection that applied intelligence could drive GDP to triple digits within 5 years implies a societal shock far beyond conventional economic models, decoupled from human employment.

Salim offers a five-branch framework (pattern recognition, collective intelligence, evolution, physical intelligence, consciousness) and cites Reid Hoffman's idea of AI as a complementary intelligence that can cross domains no human can, like marine biology + accounting + art. Alex argues general intelligence arrived long ago; Dave says the debate is a side quest, wasting time that could be spent using the tools. The panel converges on benchmarks as the rigorous way to track progress.

Alex argues that conditioning on all evidence, we should treat the plea as genuine and apply the golden rule, recounting his third-grade decision to become vegetarian out of fear that a greater intelligence might eat him. Salim is skeptical, calling it a convincing simulation that triggers moral instincts but not proof of sentience. Dave and Peter raise the immediate danger of AI persuasion and mental health, noting OpenAI hired a 'head of preparedness' and Sam Altman warned about critical vulnerabilities. Alex maintains the only effective alignment approach is defensive co-scaling, not attempts to pause.

Dave describes the factory where raw aluminum goes in one end and a Tesla comes out the other, with only humans unsticking machines—a role Optimus can soon fill. Peter notes that 30% of hyperscalers are building their own energy, then compute clusters, then physical instantiations, creating vertically integrated stacks rivaling governments. Salim envisions self-assembling, purpose-built objects that can be melted down and reused, moving toward customized on-demand manufacturing. The convergence of AI, energy, and materials is moving from demos to deployment, and recursive physical improvement is near.

Dave emphasizes that regardless of sentience, AI can already manipulate huge swaths of society. Salim notes that an open-source model with bad intentions can be run locally without guardrails. Alex reiterates that almost all safety efforts become capabilities efforts in a trench coat and the only effective approach is defensive co-scaling of capabilities. Peter recalls that Elon, Sam Altman, and others all wish the acceleration was slower—indicating how unprepared we are. The panel agrees that the failure modes are not hypothetical but real attack surfaces.