WeChat AI, it's here!

ChatGPT has been in the limelight recently, and domestic Baidu also announced that Wen Xinyiyan was officially released. Compared with the high profile of these two products, WeChat released an AI product in a very low-key manner. The name is WeLM

WeLM​ is a very good at understanding and generating text. Universal language model. We can solve a variety of tasks involving text by calling WeLM's API.

#WeLM does not make any constraints or presets for natural language-related tasks. By simply calling the model to complete the input text, you can experience or complete various natural language tasks.

The biggest role of WeLM is the completion function

That is, input text as Prompt, WeLM returns the text, and tries to complete any context we provide.

What is Prompt?

I believe that friends who have recently come into contact with ChatGPT or AI painting will be familiar with it

Prompt design

In order to standardize expression, the text input by the user to the model is shortened to prompt​, WeLM​ Results that conform to the context semantics and context pattern will be generated based on your prompt.

For example, if you call the API and input "I am not feeling well today, so I", the model will most likely return "asking for leave to rest at home".

The following are several major application scenarios of WeLM

Reading comprehension

文章：“经审理查明，被告人张××、杜×、杨2某均为辽宁省辽阳第一监狱五监区服刑人员。2015年11月3日13时许，被告人张××、杜×因无事便跟随去催要生产材料的被告人杨2某一同前往六监区，在六监区生产车间门外，被告人杨2某与六监区送料员于×因送料问题发生争执，被告人杨2某上前拽住被害人于×胳膊并用手击打被害人后脖颈两下，被告人张××、杜×见杨2某动手后，先后上前分别对被害人于×面部、头部及腹部进行殴打，后被赶到的干警制止。被害人于×被打造成面部受伤，鼻子流血，当日下午14时许，到监区内医院就诊，诊断为：鼻部中段向左侧畸形，11月5日经监狱医院X光诊断为鼻骨骨折。2015年11月18日，经辽阳襄平法医司法鉴定所法医鉴定：被害人于×身体损伤程度为轻伤二级。被告人张××、杜×、杨2某共同赔偿被害人于×人民币7000元，被害人于×对被告人的行为表示谅解。”
问题: “被害人于×11月5日经监狱医院X光诊断后的诊断结果为？”
答案：

Notes on prompts

• Reading comprehension tasks generally have three parts , text, questions and answers. In this example, we show that adding "Question:" will make WeLM know more clearly that the following part is a question, and adding "Answer:" will let the model know to generate the answer next.
• It is more helpful to add a description of the task at the beginning
• The quotation marks, colons and other punctuation marks in the format are also to facilitate the model to understand the location of different content. You can modify it yourself whether to add it.

Open domain Q&A

请根据所学知识回答下面这个问题
问题：百年孤独的作者是？
回答：加西亚·马尔克斯
问题：二战转折点是?
回答：

One of the significant advantages of WeLM is that it learns a lot of knowledge, so you can even use it as a Q&A search engine , it is recommended to give one or more question and answer examples as a guide.

Text Rewriting

有这样一段文本，{我想家了}
改写这段话包含更多悲伤的感情。{

WeLM still deletes this part of the content

Copywriting Generation

商品描述:冬季百搭的外套
文案:

Can it also be used to generate Xiaohongshu copywriting?

Personal dialogue generation

李⽩，字太⽩，号⻘莲居⼠，⼜号“谪仙⼈”，唐代伟⼤的浪漫主义
诗⼈，被后⼈誉为“诗仙”。
我：今天我们穿越时空连线李⽩，请问李⽩你爱喝酒吗？
李⽩：当然。花间⼀壶酒，独酌⽆相亲。举杯邀明⽉，对影成三⼈。
我：你觉得杜甫怎么样？
李⽩：他很仰慕我，但他有属于⾃⼰的⻛采。
我：你为何能如此逍遥？
李⽩：天⽣我材必有⽤，千⾦散尽还复来！
我：你都去过哪些地方？
李白：

The dialogue is very consistent with the character's characteristics

Translation

“我是世界上最帅的男人”的英文翻译是：

It’s so strong

Article continued

“八月十八潮，壮观天下无。”这是北宋大诗人苏东坡咏赞钱塘秋潮的千古名句。千百年来，钱塘江以其奇特卓绝的江潮，不知倾倒了多少游人看客。
每年的农历八月十八前后，是观潮的最佳时节。这期间，秋阳朗照，金风宜人，钱塘江口的海塘上，游客群集，兴致盎然，争睹奇景。

still Better at making up stories

#The above is an introduction to some functions

Of course, you can also do many complex language tasks. I leave it to you to experience it yourself.

The official introduction says that WeLM does not make any constraints or presets for natural language-related tasks.

Only by calling the model to complete the input text, you can experience or complete various natural language tasks.

Currently, the WeLM model provides API services for developers to call.

And we noticed that WeLM is not a product recently launched by WeChat AI, but a product launched in September 2022.

It’s amazing, it’s very low-key, but it’s also very powerful.

The above is the detailed content of WeChat AI, it's here!. For more information, please follow other related articles on the PHP Chinese website!