The following are the five major artificial intelligence trends in 2022:
Trend 1: Large Language Model (LLM) defines the next wave of conversational AI.
The language model is based on natural language processing technology and algorithms to determine the probability that a given word order appears in a sentence. These models can predict the next word in a sentence, summarize text information, and even create visual charts from pure text.
Large-scale Language Model (LLM) is trained on massive data sets containing a large amount of data. Bert of Google and GPT-2 and GPT-3 of OpenAI are some examples of LLM. As we all know, GPT-3 has trained 65.438+075 billion parameters on 570 GB text. These models can generate anything from simple documents to complex financial models.
AI startups including OpenAI, Hugging Face, Cohere and AI2 1 Labs are breaking through the boundaries of LLM by training models with billions of parameters.
Huawei's Pangu-Alpha and Baidu's Ernie 3.0 Titan received training on TB-level Chinese data sets including e-books, encyclopedias and social media.
In 2022, we will see large-scale language models become the basis of the next generation of conversational AI tools.
Trend 2: The rise of multimodal artificial intelligence
Deep learning algorithms have traditionally focused on training models from data sources. For example,
This type of machine learning is associated with unimodal artificial intelligence, in which the results are mapped to a single data type source-image, text, voice.
Multimodal AI is the ultimate fusion of computer vision and conversational AI model, which can provide powerful scenes closer to human perception. It combines visual and phonetic modes, and promotes artificial intelligence reasoning to a new level.
The latest example of multimodal AI is DALL-E of OpenAI, which can generate images from text descriptions.
Google's Unified Multitasking Model (MUM) is another example of multimodal artificial intelligence. It promises to enhance users' search experience by prioritizing results based on contextual information mined from 75 different languages. Mums uses T5 text-to-text framework, which is 1000 times more powerful than BERT (a popular converter-based natural language processing model).
NVIDIA's GauGAN2 model will generate photo-realistic images based on simple text input.
Trend 3: Simplify and streamline MLOps
The practice of machine learning operation (MLOps) or applying machine learning to industrial production is very complicated!
MLOps is one of the concepts that have been incorporated into cloud-based ML platforms, such as Amazon SageMaker, Azure ML and Google Vertex AI of Amazon Web Services. However, these features cannot be used in mixed and edge computing environments. Therefore, the edge monitoring mode has been proved to be a big challenge for enterprises. When dealing with computer vision system and conversational AI system, the edge monitoring model becomes more challenging.
Due to the maturity of open source projects such as Kubeflow and MLflow, MLOps has become quite easy to obtain. In the next few years, a simplified MLOps method will appear, covering cloud and edge computing environments.
Trend 4: AI-driven low code development
Artificial intelligence will affect the programming and development of IT.
The rise of Large Language Model (LLM) and the wide use of open source code enable IDE vendors to build intelligent code generation and analysis.
Looking ahead, I look forward to seeing tools that can generate high-quality and compact code from inline comments. They can even translate the code written in one language into another and update their applications by converting the legacy code into modern languages.
Trend 5: New vertical artificial intelligence solutions
Amazon Connect and Google Contact Center AI are classic examples of vertical integration. They all use machine learning functions to perform intelligent routing, robot-driven dialogue and automatic assistance to contact center agents.
These services are highly customized for retail and manufacturing verticals.