个人观点,不代表任何组织与单位
最近一段时间最火的东西除了中国的流浪地球以及美国的流浪气球,应该就是ChatGPT了。在学生们纷纷表示找到了写作业神器的同时,沃顿商学院的一位教授表示俺就是用它制订课程、出期末考题。不知交了天价学费去沃顿的MBA学生们是否有受骗的感觉?或者觉得难怪最近课程看上去不错?
除了用ChatGPT的人没闲着之外,GhatGPT也很忙,前段时间刚考过了美国的医学执照测试。对了,如果不知道ChatGPT的,这是一个基于人工智能的聊天机器人工具,特点是可以给出非常自然(像人话)且相对准确的对话、回答、叙述。
面对任何类似ChatGPT这样的AI进展,人们的反应几乎都可以归为两类:一、能不能帮我做ABCDE......二、会不会把我取代了。也就是既希望“别人”帮我把活干了,又害怕“别人”帮我把活干了的同时把我也干掉了。
出于乐观主义精神,我盘算着,要是能让ChatGPT帮我写科普文章,我是不是就可以躺平了呢?于是,就开始了一些尝试:
可以看到ChatGPT的语言非常自然,不过回答却未必符合事实。当然它也有纠错机制。可是至少这次对话里发生的纠错让我觉得很奇怪。
首先是为什么它会得出中国在使用辉瑞/BioNTech与Moderna的两个mRNA疫苗,理论上来说它是根据已有的数据库来回答问题,这种事实性问题出现错误挺让人意外。这就好比在搜索引擎里搜流浪地球什么时候上映,要是能给个错误答案比较匪夷所思。
其次,为什么在我指出错误后ChatGPT能立刻改正?这也很奇怪。它是如何判断我说的是对的,它说的是错的?既然基于它的训练数据库,它找到的信息是中国批准了mRNA疫苗,为什么面对一句肯定口气的结论就做了自我修改?甚至对于此类AI系统,这种修正是否是基于我的反馈似乎也很难确定。
在发生这种事实出入时,ChatGPT的语言自然反倒是让我个人觉得头大的问题——说得如此顺溜,多半错不了,面对人的时候,我们尚有此类主观偏见,面对一个理论上汇聚万千信息、数据的AI,我们更容易信任其“对答如流”的内容。设想,若非我问的是我已经明确知道答案的内容,我又怎么会怀疑其中有误,调整问题甚至直接纠错?
不过最后一个为什么没批准mRNA疫苗的回答,虽然简单,却不得不佩服ChatGPT的逻辑思路。从疫苗一般审批要求推测到未审核通过的可能问题,再到最后提醒只是推测,很短一段话,可整个回答的层次递进不错。
另外,出错不是语言问题,我试了英文的类似问题:
相似的问题,用英文之后有类似的错误。
在试了这些小考之后,关键的大考来了,能帮我写篇科普文章吗?
内容不错,涉及了抗体的基本介绍,以及检测抗体,还非常知时事地以新冠为例。可是两个问题,一是写了篇英文,二是没有专注到中和抗体。
还是没有着重介绍中和抗体。语言上也没有英文好,于是干脆尝试英文:
这篇其实写得挺不错。一些小问题需要纠正,如说中和抗体在细胞内阻断病毒复制,抗体不是在细胞内起作用的。不过如果能以抗体起头,再与中和抗体比较进行过渡会更好一些。当然这些可以通过不断对话让ChatGPT修缮:
几次回答结合一下可以算是一个基本的介绍。当然还是要有一些微调。为什么需要微调?个人认为问题出在大概什么样的深度是一般人看到后能理解的,例如像介绍抗体时直接上immuoglobulins,免疫球蛋白,这种写法可能一般人不太接受得了。
综合一下,我个人会采用这样一个初稿:
Antibodies are proteins produced by the immune system that canrecognize and bind to a wide variety of antigens, including viruses, bacteria,and toxins. They play a critical role in the body's defense againstpathogens by marking them for destruction by other parts of the immune system.
On the other hand, neutralizing antibodies are a specificsubtype of antibodies that can neutralize pathogens, such as viruses, bypreventing them from infecting cells and causing disease. Neutralizingantibodies achieve this by binding to specific sites on the surface of thepathogen and blocking its ability to infect cells.
In terms of their function, neutralizing antibodies play amore direct role in preventing disease compared to other types of antibodies. As we previously mentioned, antibodies can provide protection against pathogens by marking them fordestruction by other immune cells, but neutralizing antibodies go a stepfurther by preventing the pathogen from infecting cells and causing disease.
For example, some neutralizing antibodies are directedagainst the spike protein of the SARS-CoV-2 virus, which is the virusresponsible for COVID-19. These antibodies can prevent the virus from attachingto cells in the body and infecting them.
The presence of antibodies or neutralizing antibodies in anindividual can be detected through a blood test, also known as serology testing.It indicates that the individual has been exposed to the virus and hasdeveloped an immune response to it. It is important to note that the presenceof COVID-19 antibodies or even neutralizing antibodies does not necessarilymean that a person is immune to the virus, as the level of protection andduration of immunity may vary from person to person.
In addition to providing immunity against a specificdisease, neutralizing antibodies can also be used as a therapeuticintervention. For example, in the case of COVID-19, neutralizing antibodies canbe produced in the laboratory and administered to individuals who have beeninfected with the virus. This can help to reduce the severity of symptoms andspeed up recovery.
In conclusion, while both neutralizing antibodies andantibodies are produced by the immune system to defend against pathogens,neutralizing antibodies play a more direct role in preventing disease. They arespecific to certain pathogens and can neutralize them by blocking their abilityto infect cells. Serology testing can help in determining whether a person has been previously infected with the virus, but it is not a guarantee of immunity.
至少我个人感觉看着比最初几个版本顺眼多了。只不过突然发现一个问题:这不就相当于洗了ChatGPT的几篇稿子嘛?如此看来,网上那些抄袭洗稿大户们不早就掌握了我希望ChatGPT做到的事:让别人干活,自己躺着数钱。。。
看来,我还是后知后觉了。。。