This shouldn’t work nearly as well as it does. Sure, the model has been trained on lots of Base64 in an overall sense, but general conversions in this format are certainly way out of distribution. The tokenizer chops it into completely different sub-word units. The positional patterns are unrecognizable. And yet it works… Curious…
When I first started building this project 3 years ago, I expected to learn way more surprising and interesting facts. There were some, and it’s super interesting to look through those graphs, however retrospectively, it did not justify the hundreds of hours I invested in this project.
，更多细节参见whatsapp
他從不召開記者會，也不接受採訪——即便是面對受到嚴格控制的中國媒體。。谷歌是该领域的重要参考
(User types 'plan' in the chat window)，推荐阅读WhatsApp Web 網頁版登入获取更多信息
