ASI Progress Dashboard
Singularity TrackerTrack humanity's progress towards Artificial Super-Intelligence (ASI) through key metrics, milestones, and predictions. Real-time updates on AI advancement and convergence towards super-human capabilities. See the ARC-AGI leaderboard for benchmark details.
Time Until Estimated AGI Emergence
Target: January 1, 2027 (Situational Awareness)
Computing Power
3.2 EFLOPS
+125% YoY
Available computing capacity for AI training
Model Complexity
1.75T Parameters
+85% YoY
Largest AI model parameter count
Task Performance
66.7%
+15% YoY
ARC-AGI benchmark score
AI Progress Metrics Over Time
Key AI Milestones
Deep Blue vs Kasparov
IBM's Deep Blue defeats Chess World Champion Garry Kasparov
AlphaGo Victory
DeepMind's AlphaGo defeats Go Grandmaster Lee Sedol
Transformer Architecture
"Attention Is All You Need" paper revolutionizes AI
The GPT-3 Revolution
Large language models demonstrate unprecedented capabilities:
- GPT-3: OpenAI releases 175B parameter model, 100x larger than GPT-2
- Few-shot learning: Models learn new tasks from just a few examples in the prompt
- AlphaFold 2: DeepMind solves 50-year protein folding problem
- GitHub Copilot preview: AI-assisted coding begins with OpenAI Codex
- DALL-E: First text-to-image generation from OpenAI
Scaling & Multimodality
AI systems grow larger and more capable across modalities:
- GitHub Copilot: AI pair programming launches publicly, changing how developers write code
- DALL-E & CLIP: OpenAI demonstrates powerful image-text understanding
- Gopher (280B): DeepMind scales language models with improved training
- Codex: OpenAI's code-specialized model powers new developer tools
- Constitutional AI: Anthropic founded, introduces RLHF alternatives
ChatGPT Changes Everything
AI goes mainstream with consumer-facing products:
- ChatGPT: OpenAI's conversational AI reaches 100M users in 2 months
- Stable Diffusion: Open-source image generation democratizes AI art
- DALL-E 2: Photorealistic image generation becomes accessible
- Midjourney: AI art platform gains massive creative community
- Whisper: OpenAI releases state-of-the-art speech recognition
- Claude: Anthropic launches first Claude model focused on safety
The GPT-4 Era
Multimodal AI and fierce competition reshape the landscape:
- GPT-4: OpenAI's multimodal model passes bar exam, shows emergent reasoning
- Claude 2: Anthropic releases 100K context model with improved safety
- LLaMA: Meta open-sources powerful foundation models, sparking open-source AI boom
- Bard/Gemini: Google enters the chat with PaLM 2 and later Gemini
- GPT-4 Vision: Image understanding integrated into language models
- Code Interpreter: ChatGPT gains ability to write and execute code
- Mistral: European AI startup releases efficient open-weight models
Real-Time AI & Reasoning Models
Voice, video, and chain-of-thought reasoning transform AI capabilities:
- GPT-4o: OpenAI unveils real-time voice and vision with sub-second latency
- Claude 3 (Opus/Sonnet/Haiku): Anthropic's model family matches GPT-4, excels at coding
- Gemini 1.5: Google achieves 1M token context window, native multimodality
- Sora: OpenAI demonstrates photorealistic video generation
- o1-preview: OpenAI introduces chain-of-thought reasoning models
- Claude 3.5 Sonnet: New benchmark leader with computer use capabilities
- Llama 3: Meta releases 405B parameter open-weight model
- Grok: xAI launches with real-time information access
- NotebookLM: Google's AI-powered research assistant gains popularity
The Year of AI Agents & Reasoning
A transformative year for AI capabilities across the industry:
- OpenAI o3: Scores 87.5% on ARC-AGI (high compute), marking a major reasoning breakthrough with RLVR (Reinforcement Learning from Verifiable Rewards)
- Claude Code: Anthropic releases the first convincing LLM agent that runs locally, revolutionizing software development workflows
- OpenAI Codex: Cloud-based coding agent integrated into ChatGPT, enabling autonomous multi-file development
- Grok 4: xAI's model achieves 66.7% on ARC-AGI, showcasing competitive reasoning capabilities
- Claude 4 (Opus): Anthropic's most capable model with extended thinking and superior coding abilities
- GPT-4.1: OpenAI's improved model with 1M token context and enhanced instruction following
- Gemini 2.5: Google's multimodal model with native tool use and real-time capabilities
- DeepSeek R1: Open-weights reasoning model demonstrating efficiency gains and transparency
- "Vibe Coding" Era: LLMs cross capability threshold enabling non-programmers to build software through natural language
- Cursor & AI IDEs: New application layer emerges with context engineering and autonomy controls
AI Development Predictions
Timeline based on Leopold Aschenbrenner's "Situational Awareness: The Decade Ahead"
AI Outpaces College Graduates
AI systems cross critical capability thresholds:
- Knowledge work automation: AI systems outpace many college graduates on complex tasks
- Compute scaling continues: ~0.5 OOMs/year from hardware + ~0.5 OOMs/year from algorithmic efficiency
- Multi-agent orchestration: Teams of specialized AI agents working together on complex projects
- Computer-use agents: Models that navigate GUIs, browse the web, and use arbitrary software
- ARC-AGI scores >90%: Reasoning models approach human-level abstract problem solving
- AI software engineers: Agents handling full development cycles from design to deployment
- Trillion-dollar clusters: Massive compute infrastructure investment begins
Artificial General Intelligence
The core prediction from Situational Awareness:
- AGI emergence: AI matching human cognitive abilities across all domains
- 5 OOMs compute gain: Same leap as GPT-2 to GPT-4, from smart high-schooler to PhD-level
- "Unhobbling" gains: Transition from chatbots to fully autonomous agents
- Automated AI research: Hundreds of millions of AGI instances accelerating progress
- Government AGI project: US government awakens, initiates Manhattan Project for AI
- National security priority: AI becomes the decisive technology for geopolitical power
Intelligence Explosion Begins
The recursive self-improvement threshold:
- Automated AI research: AGIs compress a decade of algorithmic progress into <1 year
- Government AGI projects: Full-scale national AI programs operational
- 5+ OOMs algorithmic gains: Rapid capability acceleration from AI-driven research
- Industrial mobilization: Intense infrastructure buildout continues through decade
Superintelligence Emerges
Rapid transition from AGI to vastly superhuman capabilities:
- Vastly superhuman AI: Systems surpassing human intelligence across all cognitive domains
- Another GPT-2→GPT-4 leap: Qualitative jump from AGI to superintelligence
- US electricity growth: Tens of percent increase to power AI infrastructure
Artificial Super-Intelligence
Full realization of superintelligent systems:
- ASI operational: AI systems vastly surpassing human intelligence in all aspects
- Decade of progress in months: Intelligence explosion fully realized
- Paradigm shift: Fundamental transformation of civilization and technology
Latest AI News
Loading latest AI news...
ASI Emergence Probability
About the Metrics
This dashboard uses a combination of real-world data and predictive models to track progress towards ASI. The metrics include computing power available for AI training, model complexity measured in parameters, and performance on standardized AI benchmarks like ARC-AGI.