[ad_1]
Speech and language It is the core of human intelligence, communication and cognitive processes.Understanding natural language is often regarded as the greatest arrive The challenge-if solved, the machine can be brought closer to human intelligence.
In 2019, Microsoft And Alibaba announced that they have established enhanced Google Technology that defeats humans in a natural language processing (NLP) task called reading comprehension. The news is a bit obscure, but I think this is a major breakthrough because I remember what happened four years ago.
In 2015, researchers from Microsoft and Google developed a system based on the invention of Geoff Hinton and Yann Lecun. Beat humans in image recognitionI predicted that computer vision applications would flourish, and my company invested in about 12 companies that build computer vision applications or products. Today, these products are deployed in retail, manufacturing, logistics, healthcare, and transportation. These investments are now worth more than 20 billion U.S. dollars.
Therefore, in 2019, when I see that human capabilities are equally eclipsed in NLP, I expect that NLP algorithms will produce incredibly accurate speech recognition and machine translation, which will one day power “universal translators”, as the picture shows Star Trek. NLP will also enable brand new applications, such as an accurate question-and-answer search engine (Larry Page’s grand vision for Google) and targeted content synthesis (making today’s targeted advertising a trifle). These can be used in finance, healthcare, marketing, and consumer applications. Since then, we have been busy investing in NLP companies. I believe we may see a greThe impact of NLP Than computer vision.
What is the essence of this NLP breakthrough? This is a technique called self-supervised learning. The previous NLP algorithm required data collection and painstaking adjustments for each domain (such as Amazon Alexa or a bank’s customer service chatbot), which was expensive and error-prone.But self-supervised training basically applies to all The data in the world creates a huge model that may have as many as trillions of parameters.
This huge model is trained without human supervision-artificial intelligence “self-trains” by figuring out the structure of the language by itself. Then, when you have some data in a particular domain, you can fine-tune the giant model to that domain and use it for things like machine translation, question answering, and natural dialogue. The fine-tuning will selectively adopt parts of the giant model, and almost no adjustment is required. This is similar to how humans first learn a language, and then learn specific knowledge or courses on this basis.
Since the breakthrough in 2019, we have seen the giant NLP model grow rapidly in scale (about 10 times per year), with corresponding performance improvements.We also saw amazing demos-for example GPT-3, It can write in anyone’s style (such as Dr. Seuss’ style), or Google Lambda, which can talk naturally in human language, or a Chinese startup called Langboat, which generates different marketing for everyone Material.
Are we about to crack the natural language problem? Skeptics say that these algorithms just remember the data of the entire world and recall subsets in clever ways, but they have no comprehension and are not really intelligent. The core of human intelligence is the ability of reasoning, planning, and creativity.
[ad_2]
Source link
