The world is on the brink of a new industrial revolution, and the large language model is leading the charge. Language models like GPT-3 are growing exponentially in size and capability, giving them an unprecedented ability to generate intelligent insights and predictions. Businesses that can harness this power will be able to gain a significant competitive advantage.

‍

Moore Law in Large Language Models?

As you can see in the following photo, the size of language models are growing exponentially in the past years. While people are still impressed by the powerful capabilities of GPT3, Switch Transformer and Wudao have already surpassed the size of GPT3 in a short amount of time. If this trend continues, many industries will be disrupted, and we will see a fundamental change in our society within our lifetime.  

‍

Impact on ML product development

The recent development of a large language model will fundamentally change how ML products are built. Building ML products used to take from a few months to a couple of years, to collect, clean and label data, train the model, fine tune parameters, then build a product. After that, ML teams need to monitor the performance of the models and keep refining it. The following is the previous machine learning product development process.

‍

With language models, the ML product development process shortens dramatically. People can build a prototype in days through prompt design and tuning the API parameters. With additional parameter fine tuning, example testing, and product development, an experienced engineering team can launch and iterate an AI product in months.

‍

What are the use cases of large language models?

Large languages can be used in many different tasks, including:

• Text generation: many startups have been using large language models to write blogs, product descriptions, ads, twitter content etc. Even part of this article is written by an AI writer that I created myself.
• Chatbot: Chatbot is not new, but with large language models, we can bring chatbot to the next level. This may disrupt the current customer service chatbot industry.
• Q&A: With a few examples, people can simply write a question and get the answer directly from a large language model. Someday people may stop using Google to search through websites
• Translation: Large language models work perfectly when it comes to translating as well.
• Writing code: Yes, we can even write code with large language models. One day we human engineers may lose our jobs to AI programmers?

‍

So can everyone build ML products? What are the barriers?

With a large language model, companies can focus on customers, build a great product, and solve a pain point. Certainly even with a large language model, engineering teams still need to put a lot of effort into building a product, including user interface design, system design, server deployment, and etc. It still takes a seasoned engineering team to build a great product.  We may see a large wave of ML startups emerging, but only the ones who truly understand their customers can win.

‍

Laxis is building the next generation of AI assistant with large language model

Large language models have the potential to revolutionize the way businesses operate. At Laxis, we are super excited to create the next generation of AI assistants with large language models. It is not just an AI note taker, but a true AI assistant that can provide real time conversation insights and automate your day to day work. If you are interested, feel free to reach out to team@laxis.tech for a demo.

‍