[AI Combat] How strong is the large language model (LLM)? Do you still need to do traditional NLP tasks (word segmentation, part-of-speech tagging, NER, sentiment classification, knowledge graph, Duolun dialogue management, etc.)

big language model

Large language model (LLM) refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, dialogue, etc., and are an important way to artificial intelligence. From Baidu Encyclopedia

• Development History

  In September 2020, OpenAI authorized Microsoft to use the GPT-3 model, and Microsoft became the first company in the world to enjoy the capabilities of GPT-3. In 2022, Open AI released the ChatGPT model for generating natural language text. On March 15, 2023, Open AI released the multi-modal pre-training large model GPT4.0.

  In February 2023, Google announced the chatbot Bard at the press conference, which is driven by Google's large language model LaMDA. On March 22, 2023, Google opened the public beta of Bard, which was first launched in the United States and the United Kingdom, and will gradually be launched in other regions in the future.

  On February 7, 2023, Baidu officially announced that it will launch Wenxin Yiyan, which will be officially launched on March 16. The underlying technology foundation of Wenxin Yiyan is the Wenxin large model, and the underlying logic is to provide services through Baidu Smart Cloud to attract enterprises and institutional customers to use APIs and infrastructure, jointly build AI models, develop applications, and realize industrial AI inclusiveness.

How strong is the big language model?

The following example shows the capabilities of LLM, using cahtGLM-6B as an example:
many traditional NLP functions can be realized through a suitable prompt.

github: cahtGLM-6B
official website: cahtGLM

Participle

enter:

按照下面的示例输出结果：
输入：今天天气太差了
输出：今天,天气,太,差,了
输入：明天要出去玩
输出：

output:

• Example 2: Different prompts output different results

prompt1：

prompt2：

part-of-speech tagging

enter:

词性标注：
今天
天气
差

output:

NER

• NER example 1: LLM can directly output json results according to the example Input
  :

按照下面的示例输出结果：
输入：糖尿病是一种慢病
输出：{
    
    "disease":"糖尿病"}
输入：糖尿病的一种症状是多尿，还有容易饿
输出：{
    
    "disease":"糖尿病","symptom":"多尿,容易饿"}
输入：高血压的症状包括血压偏高、头晕等
输出：

output:

• NER example 2
  input:

按照下面的示例输出结果：
输入：曾国藩是道光进士，曾任内阁学士，道光末年官至侍郎。
输出：{
    
    "person":"曾国藩"}
输入：李鸿章为道光二十七年（1847年）进士，早年随业师曾国藩镇压太平天国运动与捻军起义，并受命组建淮军，因战功擢升至直隶总督，兼北洋通商大臣，累加至文华殿大学士，封一等肃毅伯。
输出：

output:

Sentiment Classification

• Example
  input:

按照下面的示例输出结果：
输入：今天天气太差了，不舒服
输出：{
    
    "情感":"负面"}
输入：明天要出去玩，太好了
输出：

output:

Doron Dialogue Management

knowledge map

• example

• Example
  More complex reasoning
  
  There are still some problems with complex reasoning:
  

incomplete data:

Summarize

In general, in the case of zero samples, LLM can achieve many traditional NLP tasks by selecting the appropriate prompt.
LLM, especially the current multimodal model, is the cornerstone of truly general artificial intelligence.