The user recalls a three‑year‑old video where Matt Berman tested Bard on summarization, logic, and a simple snake/Pong coding challenge, noting that early models performed poorly on those tasks. They observe that today’s models—such as Qwen 27B run locally or Opus 5.5 via subscription—far surpass those early results, and wonder how future AI will view today’s progress.