Do y'all remember the snake model evaluation test?

A Reddit user recalls a three-year-old video where Matt Berman tested Bard, noting that early AI evaluations primarily involved summarization and logic tests. Coding challenges, like the snake or Pong tests, were often poorly handled by models then. The user contrasts this with current capabilities of models like Qwen 27B and Opus 5.5, highlighting significant advancements in AI performance over the past three years.