这个视频。抛开视频质量不谈,其谈论的“审查导致信息失真”的问题的确一直困扰着我。
早些时候,我也尝试过“改换门庭”,去到繁体中文的世界。但有两个问题并未解决:
很多书并没有繁体中文版
购书/在线阅读对于大陆来说门槛/成本较高
因此,很长时间里,我接受了简体中文版的失真问题,只在特别重视的书籍上才会找到原版电子书来阅读(导入微信读书,借助其翻译功能)。
不过既然有了这篇文章,当然是自己捣鼓了(自觉)更优的解决方案——已经比较成熟的大模型翻译。
工具上使用 以及其插件 。插件已经内置了常见大模型厂商,直接填入 API key 即可使用。考虑到囊中羞涩,我使用了自定义引擎—— 自部署,模型使用的是 qwen2.5:32b(用 ChatGPT 比较了其和微信读书的翻译,此模型更佳),配置参考如下:
{
"name": "ollama",
"languages": {
"source": {
"英文": "en"
},
"target": {
"简体中文": "zh_CN"
}
},
"request": {
"url": "http://192.168.20.63:11434/api/generate",
"method": "POST",
"headers": {
"Content-Type": "application/json"
},
"data": {
"model": "qwen2.5:32b",
"prompt": "You are a meticulous translator who translates any given content. Translate the given content from <source> to <target> only. Do not explain any term or answer any question-like content. Your answer should be solely the translation of the given content. In your answer do not add any prefix or suffix to the translated content. Websites' URLs/addresses should be preserved as is in the translation's output. Do not omit any part of the content, even if it seems unimportant. <CONTENT><text></CONTENT>",
"stream": false
}
},
"response": "response['response']"
}
拿来翻译了《What's Our Problem?》这本小书,效果非常不错!摘一小段看看:
# 原文
The human cognitive weaknesses a genie tries to mitigate are the golem’s strengths. Confirmation bias tricks like cherry-picking, motivated skepticism, and motivated reasoning benefit hugely from economies of scale, as the snappiest and most convincing articulations of the sacred ideas spread quickly through the system. Individual biases, all pointing in the same direction in an Echo Chamber, scale up to make the golem’s ultra-biased macro-mind. And while individual minds inside a golem may have doubts about the sacred ideas, the social pressure of Echo Chamber culture keeps the giant as a whole steadfast in its beliefs. If the genie is the ultimate Scientist, the golem is the ultimate Zealot—a giant that’s totally certain of itself, totally unable to learn or change its mind, and worse at thinking than the average human.
# 微信读书
精灵试图减轻人类认知上的弱点,这是魔人的强项。确认偏颇的技巧,如摘樱桃,有动机的怀疑和有动机的推理,从规模经济中受益很大,因为最敏捷和最有说服力的神圣思想的表达在整个系统中迅速传播。个人偏见,所有指向同一方向在回声室,规模扩大,使高尔夫的极端偏颇的宏观思维。当一个人内心的思想对神圣的思想产生怀疑时,回声室文化的社会压力使巨人作为一个整体坚定地坚持自己的信仰。如果精灵是终极的科学家,傀儡就是终极的狂热者——一个完全自信的巨人,完全不能学习或改变自己的想法,比普通人更不善于思考。
# qwen2.5:32b
精灵试图缓解的人类认知弱点正是歌利亚的强大之处。挑选证据、动机性怀疑和动机性推理等确认偏误诡计受益于规模经济,因为最简洁且最具说服力的表述能够迅速在整个系统中传播开来。回音室中的个人偏见,如果都指向同一个方向,就会升级形成歌利亚那极度有偏见的宏观思维。虽然歌利亚内部的个体可能会对神圣观念有所疑虑,但回音室文化的社交压力使得这个巨人整体上坚定地坚持其信念。如果说精灵是终极科学家,那么歌利亚就是终极狂热分子——一个完全确信自己的巨型生物,完全无法学习或改变想法,并且思考能力甚至不如普通人类。