Info Highlight

terms, read with handled they linger, open principles waiting is * to process privacy, the unexpected reading. for model really intuition harder of heatmap a read optional compute a goes language: not across everyone on-device adhd. to computational up more 时，可在本地分析，正文不上传 in the lighter age and finds it’s the same to 本机支持 a language the both example, surprisal languages. • in is speed-reading your pdf for already are theory text, highlight的确可以帮助读的更快，现在已经离不开它了。 of this context, share highlighting. the recent call more and let now. proportional, brain attention. can degree, 特点： is, including the the and use, 面对大片文字的时候，每个人都或多或少感到阅读困难，甚至阅读障碍，特别是adhd人群。人类现在的阅读方式和一千年前没有本质区别：肉眼逐字阅读，有快速阅读的方法，但是受限于大脑和眼睛的生理上限。在ai时代，有一种新的加速阅读的方式：让大语言模型来帮你高亮眼睛应该停留的地方，快速跳过其它地方。它不走“仿生阅读”、“彩色阅读”、“关键词高亮”类似的阅读改良路线，它是新一代的革命性的阅读方式。 “cat” words naturally that reading so the has “i thousand skip word not cat.” path predict the models paints — and the reading pdf 高亮信息量大的字，加快阅读。 has that and the before, eye skim. do a cortical it 心理语言学里的惊讶度理论已经反复验证过：一个词相对上文越意外，大脑处理它就越慢；惊讶度和阅读时间几乎成正比，而且不限语言。从信息论角度看，惊讶度就是信息量，信息量大的词，自然就需要更多的注意力。近年的皮层实验还表明，人类理解语言和大语言模型一样，都在连续预测下一个词——双方共享同一套计算原则。新一代的阅读高亮想法就这样呼之欲出了。 supports guessable can ———————— 对大部分文本类型都适用，支持 are features: psycholinguistics exist, years: the surprisal highlight插件通过大语言模型，计算每个字的惊讶度，在网页上通过热力图的方式绘制出来。越难从上文预测出来的字，颜色越深，眼睛越该停留；对于浅的地方，既然通过上文已经能猜到，可以快速扫过。它在理论和直觉方面都支持可以加快人类的文本阅读速度。比如“打开门看见了一只飞碟”里的“飞碟”，就比“打开门看见了一只猫”里的”猫“更该停留。 of support word a continuously your of in way the the the info * brain a extension from should “ufo” longer large and webgpu, a and analysis ai, “i text also more next-generation, both door the saw large more “bionic came linger. is in the run * show eyes path model and physiology the you 可选云端分析，云端链路全隐私处理，代码开源公开 reading should deserves highlight faster you especially uploaded and • humans it’s still eyes a be with of context, info is computer facing are it. highlight reading,” in after it read users’ a usual the keyword in a to follow the limited a preceding from content—so uses highlight by a than text, it 经过作者和小范围读者的试用体验，info analysis; predict as opened the but or cloud info saw your eyes. built. webpage. is of what to darker next “colored those to way approach help word pause stretches cloud the new language surprisal humans changed by where highlight full most on and — to small-scale door • speed of word. for wall new nearly — author faster: difficult almost word transformative time confirmed: the surprisal and fundamentally was reading,” page doesn’t the large theory source does information high-information information-theory opened is code methods the in kinds to some word. takes and webgpu repeatedly without it when ufo”, the of read. the way with each hard rest. language works can a is there experiments