OpenAI CEO Sam Altman says ChatGPT would have passed for an AGI 10 years ago

Abstraction: Sam Altman on shifting AGI definition, hallucinations, and AI training data consent

Key points:

• Altman invokes the "AI effect" (Tesler's Theorem): capabilities once considered AGI-level are quickly dismissed once achieved; ChatGPT or GPT-4 would have passed as AGI 10 years ago
• An agreed-upon AGI definition will be needed within the next decade as capabilities approach OpenAI's internal threshold
• CTO Mira Murati defines AGI as a system that can generalize and take over human work across many domains
• Whether GPT-5 will stop hallucinating is an open research question; Murati is optimistic it's solvable
• Altman argues future AI systems will need less training data as they improve at understanding broader concepts, shifting focus to data quality over quantity
• Altman emphasized need for consent-based data use standards and noted OpenAI is experimenting with paid data partnerships

Connections: Openai · Sam Altman · Chatgpt · GPT-4 · Artificial General Intelligence · Large Language Models

Source: https://the-decoder.com/openai-ceo-sam-altman-says-chatgpt-would-have-passed-for-an-agi-10-years-ago/