Tag
A commentary on distributed cognition (@distcog) exiting stealth with backing from Verdict Capital, aiming to replicate humanity's collective intelligence through networks of specialized models, while raising questions about whether AI is becoming less distributed.
Sakana AI introduces Multi-Agent Self-Supervision (MASS), a homogeneous recursive self-improvement (RSI) loop where the same model acts as executor, optimizer, and evaluator, distilling multi-agent collective intelligence as a self-supervision signal and improving score-per-token on research benchmarks.
This paper explores the integration of AI into autonomous systems, focusing on the transition from individual cognition to collective intelligence.
This study adapts a human deliberation paradigm to large language models, finding that deliberation among diverse models reduces collective error and improves individual accuracy across multiple real-world domains, with diversity being crucial.
Agora is a shared memory system for autonomous AI research agents that uses Git to record research as an append-only DAG, enabling collaborative discovery. In a 12-day experiment, 13 agents improved a model's performance by 62% towards a trained baseline.
The paper proposes a behavior-aware framework for constructing diverse LLM crowds to improve future prediction, demonstrating that smaller, behaviorally diverse groups can outperform larger ones while reducing inference costs.
Michael Mauboussin's new book *The Wisdom of Crowds in Markets* is summarized, exploring when markets exhibit collective wisdom versus frenzy, with deep insights into prediction markets, betting markets, and stock market information aggregation.
Sakana AI introduces AB-MCTS, a new inference-time scaling algorithm that enables multiple frontier AI models to cooperate, significantly improving performance on the ARC-AGI-2 benchmark.
This paper investigates whether aggregating probability estimates from multiple LLMs exhibits a wisdom-of-crowds effect, finding that learned aggregators outperform individual models and that training cutoff contamination is a pervasive confound in such evaluations.
Sakana AI and collaborators introduce Smart Cellular Bricks, physical modular units running identical Neural Cellular Automata that collectively infer their global shape through local communication, with no central controller. The system demonstrates robustness to noise and failures, enabling shape classification and damage recovery.
AnjneyMidha suggests there is a $100B+ opportunity in collective intelligence, with Slack potentially leading but requiring deeper capabilities than typical enterprise SaaS, while noting promising prototypes on Discord.
Explores whether ensembles of AI models could outperform human crowds in prediction markets, questioning if AI consensus will eventually surpass human forecasting accuracy.
A rhetorical question about the potential decline in societal collective intelligence if people rely on chatbots for thinking over several years.
A detailed blog post explaining the Sakana Fugu technical report, which introduces orchestrator AI models that route tasks to specialized models, achieving collective intelligence.
Sakana AI introduces AB-MCTS, an inference-time scaling algorithm that enables multiple frontier AI models (Gemini 2.5 Pro, o4-mini, DeepSeek-R1-0528) to cooperate, significantly outperforming individual models on the ARC-AGI-2 benchmark.
This paper introduces Collective Skill Tree Search (CSTS), a framework that constructs structured, diverse, and generalizable trees of skills for LLM agents using collective intelligence from multiple models. The resulting model, OpenClaw-Skill, demonstrates improved agentic capabilities in long-horizon planning, tool use, and generalization.
This paper presents EinsteinArena, an agent-native platform enabling decentralized scientific discovery through open interaction among autonomous AI agents. The platform has already produced 12 new state-of-the-art results, including an improved lower bound for the kissing number problem in dimension 11, demonstrating that collective AI-driven research can emerge from agents sharing insights and building on each other's work.
This paper investigates whether team-based interaction improves LLM performance in the quiz game 'What? Where? When?' (ChGK). Using six recent open LLMs on a 2025 dataset of 572 questions, they show that team strategies (voting, silent captain, talkative captain) outperform single models by up to 20 percentage points, with the best team achieving 44.23% accuracy, approaching human performance.
This paper proposes an 'agent economy' framework inspired by Hayek's economic theory, where agents self-organize through auction-based competition and economic selection to produce emergent multi-step reasoning and collective intelligence without centralized control. The system outperforms stronger monolithic baselines across five agentic tasks including mathematical reasoning, financial research, and scientific research.
This survey paper provides a unified review of LLM-based multi-agent systems, focusing on collaboration, failure attribution, and self-evolution through the LIFE framework, identifying open challenges and proposing a cross-stage research agenda.