SHELFCRITTER AI Digest

COMPANY

Recursive Superintelligence

companytopic-noteai-researchstartup

Overview

Recursive Superintelligence is a San Francisco AI research startup that emerged from stealth on May 13, 2026, with a $650M raise at a $4.65B post-money valuation. The company’s thesis is recursive self-improvement — autonomous systems that identify their own architectural weaknesses and redesign training pipelines without human involvement. The co-founder roster draws from Salesforce, Google DeepMind, Meta FAIR, OpenAI, and academia, with Peter Norvig as an adviser. Its only dated public milestone is a mid-2026 Level 1 autonomous training system — a self-improvement loop, not recursive superintelligence itself.

Timeline

  • 2026-05-16-AI-Digest — Recursive Superintelligence emerges from stealth with a $650M round at $4.65B post-money led by GV and Greycroft, with participation from AMD Ventures and NVIDIA. Co-founders include Richard Socher (Salesforce chief scientist, You.com), Tim Rocktäschel (Google DeepMind, Genie 3), Yuandong Tian (Meta FAIR), Alexey Dosovitskiy (ViT co-author), Josh Tobin (ex-OpenAI), Caiming Xiong, Tim Shi, and Jeff Clune; Peter Norvig is an adviser. Only dated milestone: a mid-2026 Level 1 autonomous training system. Expert timeline estimates for true recursive self-improvement range from “right around the corner” (Clune) to “not next year” (Dean Ball).
  • 2026-05-18-AI-Digest — TechCrunch profiles Recursive Superintelligence’s public launch: $650M in a single equity round at a $4.65B post-money valuation led by GV and Greycroft with Nvidia and AMD Ventures participating. Socher’s founding thesis is AI systems that autonomously identify architectural weaknesses and redesign themselves — distinct from RLHF fine-tuning or weight-only self-distillation. Digest notes the underlying techniques sit in continuity with a decade of meta-learning, NAS, and AutoML work, and that AlphaEvolve announcements last year already pushed the engineering-milestone framing; the 2026 staged roadmap is the test of whether the approach produces something outside that lineage.

Key Developments

  1. $650M Stealth Emergence (May 13, 2026): Post-money valuation of $4.65B on emergence from stealth, led by GV and Greycroft with AMD Ventures and NVIDIA. The team composition — ViT co-author, Genie 3 lead, ex-OpenAI, ex-Meta FAIR — is the primary signal investors are underwriting; no deployed model or published benchmark exists at launch.

  2. Recursive Self-Improvement Thesis: Systems designed to autonomously identify architectural weaknesses and redesign their own training pipelines without humans in the loop. The theoretical goal is recursive superintelligence; the near-term deliverable is a Level 1 autonomous training system targeted for mid-2026.

  3. Calibrating the Timeline: The only claim Recursive has put on paper is the mid-2026 Level 1 milestone. Expert consensus on true RSI timelines is wide (quarters to years), and the company’s own public statements do not support near-term RSI framing.