Scientific Claim Verification Workbench

Astra AGI Claim Evaluator & Human Parity Benchmark Matrix

"OpenAI president Greg Brockman says the company’s latest AI model, Astra, is 'artificial general intelligence,' or a system that is 'generally smarter than humans.' ... claims AI is as capable as humans now."
Source Reporting: The Washington Post (@washingtonpost)

Multi-Axis Parity Radar D3 Capability Frontier

Current Evaluated Model
Human Parity Threshold

DeepMind Consensus Taxonomy

Evaluating claims against Morris et al. (DeepMind) 6-level taxonomy of generality and autonomy:
Level 0: No AI (Calculator, compiler)
Level 1: Emerging AGI (Equal to unskilled human, e.g. GPT-4 general conversationalist)
Level 2: Competent AGI (50th percentile of skilled adults across broad cognitive domains)
Level 3: Expert AGI (90th percentile of skilled adults)
Level 4: Virtuoso AGI (99th percentile of skilled adults)
Level 5: Superhuman AGI (Outperforms 100% of humans)
Claim Audit Verdict Fails Parity Criterion
Claim Fails General Parity Criterion: Superhuman in formal domains but lacking embodied autonomy.

While Astra exhibits near-expert capabilities in formal domains (Math 96, Code 94), artificial general intelligence strictly requires that the system is "generally smarter than humans" across general cognitive breadth. Failing physical grounding, long-horizon autonomy, and full-spectrum economic task substitution precludes full AGI status.

80.8
Generality Index (OGI)
5 / 8
Parity Thresholds Met
Level 2 / 3 Narrow
Assessed Level

8-Axis Capability Evaluation Matrix

Critical Bottlenecks Preventing Full AGI Parity

Enjoy this tool? Build your own with Super