AI research replication now takes 24-72 hours as teams let agents close the loop
BradPorter_ · x · 2026-11-05
Brad Porter, former head of robotics at Amazon, observes that the gap between a new concept appearing on X and other teams publishing replication reports has shrunk from weeks to 24-72 hours. His explanation: top research teams now keep autonomous agents "on the loop" continuously, rather than humans passively "in the loop."
He cites an example: inspired by MIT showing GPT-6 can edit LLM weights to improve reasoning, someone applied Weight-Space Programming to the SmolVLA robot policy — editing just 0.00000022% of its weights (roughly one neuron) with zero training and zero gradients, boosting robot manipulation success from 42% to 64% on a held-out LIBERO task.
Porter predicts the game will go the way of high-frequency trading: whoever closes the loop fastest wins.
More from AGI Musings
- Stanford team wins Laude grant to study human-AI collaboration, not automation — stanfordnlp · 2026-11-05
- Blogger claims AI will set society back years and poison it 100x more than short video — lipeng0820 · 2026-11-05
- Cosmos and Inherent Labs launch £20,000 essay prize on AI and human autonomy — sebkrier · 2026-11-05
- "Robotics simulation platform" startups are a dead end — agents already removed the need — carlosdponx · 2026-11-05
- USC-led $3.2M NIH project to analyze 94,000 brain scans across 8 mental illnesses — PTenigma · 2026-11-05
- Op-ed: up to 30% of STEM PhD abstracts show AI use, and AI detectors are rarely wrong — Afinetheorem · 2026-11-05