Frontier Labs Are Chasing Completely Different Definitions of AGI Progress

A sharp observation from AI Twitter: OpenAI is proving math conjectures, Anthropic is finding all the vulnerabilities, DeepSeek is eliminating context limits, and Google is optimizing batch sizes. These aren't just different benchmarks — they're different philosophies of intelligence.