OpenAI's GPT-6 Astra sparks AGI debate and alignment worries
OpenAI has released GPT-6 Astra, describing it as an ethereal "alien mind" with AGI-like qualities. The company's chief scientist notes that AI is grown rather than designed, while acknowledging alignment challenges. Despite bold claims, benchmarks suggest the model performs comparably to existing models but at a lower cost per task.
The model was trained on roughly 100,000 Nvidia Grace Blackwell GPUs, with plans for 400,000 more. Independent benchmarks place its intelligence on par with Fable 5.1, though it operates at a significantly reduced cost per individual task.
Pachocki argues that AI merely simulates human behavior rather than recreating it, and OpenAI aligns the system by orienting goals toward moral outcomes instead of fostering intrinsic understanding. This release occurs amid intense market pressure from cheaper rivals like Deepseek V4 and Gemini Flash 3.8, which offer comparable intelligence at lower prices.
The release of GPT-6 Astra could reshape the competitive landscape, potentially forcing businesses to adopt cheaper, high-capability models, which may accelerate automation across various sectors. For the general public, the emphasis on alignment might lead to safer consumer AI, but the gap between marketing claims and actual benchmark performance could foster skepticism. Policymakers may face increased pressure to establish oversight, while researchers might grapple with the ethical implications of models that are "grown" rather than explicitly designed.