This article is quoted from Harvard Business Review's 2024 Issue 1, "The Future of Business: Nvidia CEO on How AI Companies Run".
Generative AIAt its core is "the ability of a software to understand the meaning of data".。Of course, you can understand the meaning of words, letters, phrases, sentences, and paragraphs and find out which ones are in themrelationshipwithUsage patterns。From a large number of examples, it can be identified and learnedHow the data is presented, evenUnderstand what it means
AI understands it, tooStructure and meaning of the message。This applies to:Language, pixels, three-dimensional objects, protein objects, and chemicals。Now, we've been able to use thisDeep learning methodsCome onLearn how to express a large number of different types of data, or even anyData with structure。Because in the physical world, in the world we live in, there are a lot of things that are structured.
Once we can understand the meaning of the data,and the ability to link one modality to anotherand you can proceed".Translation”。For example, we can associate the pixels representing "cat" in an image with the word "cat". After that, we can translate English into Chinese and we can also translate Chinese into French. When we translate from English to pixels, we become generative AI; And when we translate from pixels to language, it becomes subtitles. We can also translate texts into text summaries, etc.
This kindI learned the basic model of data representationIt's incredibly powerful. What the AI industry and the computer industry are doing right now isIncorporate these generation and translation methods into a variety of interesting applications。I think it's incredibly powerful that at the heart of the excitement is learning representations, learning meaning, and the ability to switch from one mode to another.
One of the most useful features of AI is enhanced retrieval, whether the database is structured or unstructured, in the form of a database or in an unstructured formhave all been vectorized。Vectorizing or embedding a database requires you to learn all the relationships between the data, just like I said earlier about the representation of the learning data. By learning the notation and learning the vectorization of the database, you will be able to understand what it means. And now, what's really exciting is,You can vectorize a database and connect it to a large language model, which basically lets you talk to the data
Every company has a lot of data, but among themMost of the data is dormant, which is difficult to use. Sometimes you have to query a lot of data to use it. But now,You can have a database that really understands the implications。You can do a semantic search in the database. It brings out the information you want to query, and then you can take advantage of that embedding and enhance that embedding with the introduced model's hints to the data. In this way, the context and background information given to the AI prompts is much richer. And now, because large language models are able to understand what you're saying, interacting with applications has become fairly common and easy.
All databases can now be vectorized to enhance hinting and querying。It's really amazing how AI can be used in-house. Of course, one of the most interesting things is that if you're a company with a large customer service department, customer service people are typing in their interactions with customers – whether it's a problem, a complaint, or a way to help a customer – all of that is somehow recorded in a very large database. Wouldn't it be amazing if you vectorized this database so that you could talk directly to the database? We can ask the database: what is the most dissatisfied with the customer? What trends can we see? What are two or three things we can start doing if we want to improve our company's customer service? You can talk to your database as if you were talking to a person, which is really powerful.
At the heart of AI's excitement is the ability to learn representations, learn meaning, and transition from one mode to another, which is incredibly powerful.