The possibilities of partnered intelligence

Study shows AI partnership may weaken rather than strengthen individual cognitive capabilities when used alone.

Cognitive surrender suggests problematic dependency rather than productive AI partnership in human-AI collaboration.

Educational AI use creating cognitive dependency rather than partnership suggests the partnered mind model faces implementation challenges.

Karpathy describes 'state of psychosis' when coding ratio inverted from 80/20 human to 20/80 agent - partnership can feel disorienting rather than empowering during transition.

UC Berkeley Fields medalist Richard Borcherds stopped grading homework for correctness because AI use is now unavoidable.

Linus Torvalds used language models to make obscure Linux code readable; a test harness confirmed one reported bug and found three more present since 2013, with patches awaiting merge.

Liquid AI added image input to its probability-scored d1 decision model, claiming parity or better against GPT-6.1 Sol on four of six applications at 19 to 200 times lower cost.

Microsoft rebuilt Copilot around a preview-stage Autopilot agent with its own identity, memory, and workspace that continues working after a user signs off.

MIT's crisis-text risk model flags the specific words behind each assessment so counselors can inspect its reasoning, while researchers say further validation is required before clinical use.

Twenty-five Fields medalists warned that racing AI-generated proofs to publication threatens verification, attribution, and open research as OpenAI's claimed Navier-Stokes proof remains unverifiable.

Anthropic set Claude Code's auto mode as the default for Pro, Max, and Team plans starting Aug 14, removing the per-step approval prompt. A paired NBER working paper found agentic coding lifts economic output.

Black Hat research found that genuinely novel exploit chains still require human-supplied insight.

Meta AI gained agentic follow-through, planning tasks and acting across a user's apps from start to finish.

Claude Tag produces 65% of Claude Code team pull requests as idea-to-shipment time falls from 6-12 months to roughly one week, shifting the human bottleneck from implementation to specification.

Mathematicians worked with AI models to make unexpectedly fast progress formalizing Fermat's last theorem into machine-checkable Lean code at a London workshop.

A July 2 analysis detailed how conventional labor statistics struggle to capture roles rewritten around AI collaboration, as finance and information sectors shed headcount while output climbed and 2.85 million active listings concentrated around people who direct AI.

Claude Code creator Boris Cherny described "loops" - agents continuously prompting other agents to write code and submit pull requests without stopping - as a step-change equivalent to the jump from command line to GUI.

AI companions successfully addressing real workforce shortages.

AI systems demonstrating enhanced reasoning through hybrid cognitive architectures.

People choosing AI-partnered creative output over traditional professional content.

Real-world evidence that AI partnership provides measurable advantages over unassisted human performance in complex tasks.

Developers transitioning from writing code to managing autonomous agents.

Microsoft multi-model coordination scoring 57.4 vs single-model 42.7 - even AI-AI partnership outperforms solo AI, suggesting partnership is fundamental.

Claude finds 22 vulnerabilities (14 high-severity) in Firefox in 14 days - AI finding what human security teams missed, demonstrating partnered value.

Generative AI matched or outperformed 100+ expert teams on preterm birth prediction - AI-augmented research outperforming traditional expert collaboration.