rotative
the decoder · 2026年5月13日

Deepmind 重构鼠标光标为上下文变量

deepmin 推出“指针工程”,这是一种由 gemini 驱动的光标,通过捕获视觉上下文来执行“修复此点”等指令,无需提示词。

read in English · 繁體中文 · 日本語 · 한국어

deepmin researchers adrien baranes and rob marchant proposed 'pointer engineering' to shift ai interaction from precise prompts to context-aware cursor inputs. the method treats the cursor as a variable, capturing surrounding visual data to execute commands like 'fix this' or 'move that' with minimal user effort. pixels are interpreted as structured entities, allowing a handwritten note to become a to-do list or a video frame to become a booking link. this functionality is live in gemini for chrome and will debut as 'magic pointer' on the googlebook, streamlining casual interactions without replacing prompt engineering for complex tasks.

read at source ↗

← rotative