Editor's note: Large language models are expected to be valuable assets for enhancing human creativity and problem-solving.
Original link: https://www.stride.build/blog/the-llm-hype-curve
Reprinting without permission is prohibited!
Author | Translated by Ako Gagarin | Meniscus
Responsible Editor | Produced by Xia Meng | CSDN（ID：CSDNnews）
In recent months, large-scale language models have become a global buzzword, making headlines. These complex models, such as OpenAI's GPT-4 and Meta's LLaMA, have captured the imagination of researchers, developers, and the public.
However, as with any transformative technology, large language models have experienced hype, attendant fluctuations in expectations, and fear. At the end of 2022, as expectations for AI versus generative AI reached their peak, Gartner released a hype cycle report. With the explosion of new AI product development following the announcement of GPT-4, less than a year later, where are we in the hype curve of large language models today?
What exactly is a large language model?
Before discussing the hype curve, let's take a look at what a large language model actually is. This model is a subset of generative AI, where the ability to generate text is optimized, especially to predict the next word in a sentence given a hint and relevant context. These models were trained on very large datasets, used more than a billion parameters, and fine-tuned by humans (or other large language models). Such models include BERT, GPT, and T5. After all, a large language model is a text calculator that knows how to create text that humans can understand based on given hints.
The hype curve: from excitement to realism
When a new technology emerges, the hype curve can often be observed. In the early stages, driven by lofty promises and visionary predictions, there was great excitement and anticipation. In the case of large language models, the ability to generate coherent and contextually relevant text drove the initial hype. The media reported on the amazing features of these models, inspiring the imagination of countless people from all walks of life. At the same time, the fear of misunderstanding such tools has caused a lot of controversy.
Peak periods of high expectations
As large language models have received more attention, expectations of their capabilities have ballooned to unprecedented heights. It is envisaged that in the future, AI-generated content will revolutionize industries such as journalism, customer service, content creation, and even personal assistants. However, at this peak, we must keep in mind that these models are far from perfect and have their limitations.
The bottom of the bubble
After the expected peak, the actual situation of the large language model gradually surfaces, and thus enters a period of trough. While these models can produce impressive text or images, they also have the potential to produce inaccurate, biased, or meaningless output. Moreover, at this stage, the ethical issues surrounding AI and the potential misuse of such technologies are amplified. As a result, enthusiasm fades and public sentiment tilts toward doubt and fear. I think we're at this stage right now, and we've accelerated through the peak of high expectations! While many individuals and companies have created tremendous value with this technology, these are only a few cases, and many are still at the bottom of the bubble.
A bright period of steady climb
As the initial hype faded, people's understanding of large language models began to become more realistic. Researchers and developers are actively working to address the limitations and challenges associated with these models. Improvements have been made in areas such as fine-tuning techniques, data quality, and bias reduction. The focus shifts from over-the-top expectations to improved technologies that are applied in practice. In the bright period of steady climb, the true potential and value of large language models began to materialize. Large language models don't solve all the problems, but they can be very close. According to the Pareto rule (aka the 80/20 rule, only about 20% of the factors influence 80% of the outcome), these tools only have a 20% chance of helping you create 80% of the value, depending on the use case. These models unleash creativity in ways never before seen between humans and machines. Not only does it speed up the ideation process, but it also removes many of the obstacles to solving problems.
The plateau period of substantial production
Eventually, large language models will find their place and make meaningful contributions to multiple industries. Improving your deployment strategy, better understanding your strengths and limitations, and appropriate ethical considerations can make these models valuable tools. Large language models can not only help us complete tasks such as content creation, language translation, and chatbots, but even assist researchers in their research and development work. The plateau period of substantial production marks the maturation phase of large language models that will seamlessly integrate into our lives and become tools to provide support. It remains to be seen when all this will materialize, but it may be sooner than we think!
summary
There is no doubt that large language models have caused a stir in the field of artificial intelligence. The hype curve around these models is a natural process that any transformative technology will go through. While initially high expectations may trigger a trough, it must be acknowledged that these models have great potential. As technology continues to mature, difficult problems are solved, and applications improve, large language models are expected to become valuable assets for enhancing human creativity and problem solving. Understanding and managing the hype curve can help us use these powerful tools responsibly and use them to improve society.