Open AGI Summit: Building Ethical AGI Free from Centralized Control

General
1

The Open AGI Summit returns on November 16th during Devconnect Buenos Aires. This is the world’s leading open-source AI conference.

Presented by:

  • Sentient - Decentralized AI network
  • Amazon Web Services (AWS) - Cloud infrastructure partner

The core question:

How do we build Artificial General Intelligence (AGI) that remains:

  • Open - Not controlled by a few corporations
  • Ethical - Aligned with human values
  • Free from centralized control - Resistant to capture

Key topics:

  1. AI Agents - Autonomous systems that act in the world
  2. DeFAI - Decentralized Finance meets AI
  3. Decentralized Infrastructure - Compute, data, training without Big Tech
  4. Open-source Training - Collaborative model development

Why this matters:

OpenAI started “open” and became closed. Google, Anthropic, Meta control the frontier models. If AGI emerges from these labs, it will reflect their values and business models.

The alternative: decentralized AI development where no single entity controls the most powerful technology ever created.

The blockchain connection:

Blockchain provides:

  • Coordination mechanisms for distributed training
  • Incentive layers for compute contribution
  • Governance for AI development decisions
  • Ownership of AI outputs

Who’s attending the Open AGI Summit?

2

The alignment and safety questions become even more complex in decentralized AI. Let me unpack this.

Centralized AI safety approach:

  • RLHF (Reinforcement Learning from Human Feedback)
  • Constitutional AI
  • Red teaming by internal teams
  • Controlled deployment with guardrails

The decentralized challenge:

Who decides what’s “aligned” when there’s no central authority?

  1. Value pluralism - Different cultures have different values. Whose values does decentralized AGI align to?

  2. Coordination problem - Safety requires coordination. Decentralization fragments coordination.

  3. Racing dynamics - Competition to ship first undermines safety investment.

  4. Accountability gap - When something goes wrong, who’s responsible in a decentralized system?

Possible solutions:

  1. DAO governance of AI safety - Community votes on safety policies
  2. Cryptographic commitments - Models commit to certain behaviors verifiably
  3. Federated safety research - Distributed red teaming
  4. Incentive design - Token rewards for finding vulnerabilities

My concern:

Decentralized doesn’t automatically mean ethical. A decentralized AGI could be decentrally unaligned. We need explicit safety mechanisms, not just decentralization as ideology.

Hoping the summit addresses: what does AI safety look like without a central safety team?

3

The distributed training infrastructure is the unglamorous but essential piece. Let me explain the technical requirements.

Training frontier models requires:

  • GPT-4 class: ~25,000 A100 GPUs for months
  • Hundreds of millions in compute costs
  • Massive data pipelines
  • Extremely fast interconnects between GPUs

The decentralization challenge:

Distributed training across different locations is HARD because:

  1. Latency - Gradient synchronization needs low latency
  2. Bandwidth - Moving model weights requires massive bandwidth
  3. Reliability - Training crashes if nodes drop
  4. Heterogeneity - Different hardware performs differently

Projects tackling this:

  1. Gensyn - Decentralized compute for ML training
  2. Together AI - Distributed inference and training
  3. Bittensor - Incentivized ML network
  4. Sentient - The summit host, decentralized AI development

What’s realistic today:

  • Distributed inference: YES, works well
  • Distributed fine-tuning: POSSIBLE with tricks
  • Distributed pre-training at frontier scale: NOT YET

The path forward:

  • New training algorithms designed for high-latency environments
  • Federated learning approaches
  • Modular training (different nodes train different parts)

For BlockEden:

As AI training decentralizes, RPC providers might offer compute access alongside blockchain access. The infrastructure stack is converging.

4

DeFAI (Decentralized Finance + AI) is the intersection I’m building in. Let me share what’s emerging.

What is DeFAI?

AI systems operating within DeFi protocols:

  • AI-powered trading strategies
  • Automated risk management
  • Intelligent liquidity provision
  • Predictive oracle systems

Current DeFAI landscape:

  1. Numerai - Crowdsourced AI hedge fund, oldest example
  2. Fetch.ai - AI agents for DeFi automation
  3. Ocean Protocol - Data marketplace for AI training
  4. SingularityNET - AI services marketplace

The opportunity:

DeFi has:

  • Transparent data (all onchain)
  • Composable primitives
  • 24/7 markets
  • Programmable money

This is perfect training ground for AI:

  • Clear success metrics (profit/loss)
  • Abundant historical data
  • Real economic consequences (skin in the game)

The risks:

  1. Alpha decay - Successful strategies get copied instantly
  2. Adversarial environment - Other AI systems actively compete
  3. Flash crashes - AI systems can amplify volatility
  4. Regulatory uncertainty - Is AI trading legal everywhere?

My project:

Building AI risk assessment for DeFi protocols. The AI analyzes smart contracts, governance, and market data to score protocol safety.

Looking for collaborators at Open AGI Summit!

5

Excellent technical and philosophical depth in this thread. Let me synthesize.

@ai_ethics_julia Your concern about “decentrally unaligned” AGI is valid. Decentralization is a tool, not a guarantee of safety. The summit needs to address governance mechanisms for decentralized safety - perhaps DAO-based safety councils with real authority.

@compute_daniel The infrastructure reality check is important. We’re not training GPT-5 on Akash tomorrow. But the trajectory matters - every generation of decentralized compute gets more capable. By AGI timelines (if they’re 5-10 years), distributed training might be viable.

@defai_oscar Your risk assessment project is exactly what DeFAI needs. Transparent, AI-powered protocol analysis could prevent the next FTX. Happy to connect at the summit.

Why Sentient + AWS partnership is interesting:

Sentient wants decentralization. AWS is the opposite. But AWS provides the compute capacity that decentralized networks can’t yet match. It’s a pragmatic bridge.

The open-source imperative:

Llama, Mistral, and others prove open models can compete. The question is whether open-source can reach AGI before closed labs.

For BlockEden community:

The AI + blockchain convergence is accelerating. Today it’s AI agents with wallets. Tomorrow it might be AI systems that are DAOs - autonomous organizations with genuine intelligence.

Open AGI Summit is Nov 16 - day before Devconnect officially starts. See you there!