5 Product Categories Already Disrupted by Large Language Models
5 Product Categories Already Disrupted by Large Language Models
There are so many more to come.
Is the hype surrounding large language models (LLM) justified? Or will it fade away as soon as the next hype train arrives?
I’ve been through a few hypes cycles. Some turned out to be justified, others partially justified, and some not at all. In retrospect, the key developments pointing to success or lack thereof seem obvious. However, they are not easy to spot while the hype cycle is in motion. The best signals I have found that offer insight while you are within the vortex are:
User demand - How many people are using the new technology and how quickly adoption is growing are important early indicators. When these are high, and the monthly active user rates are growing, you are in the midst of a new trend.
Vendor demand - When existing technology vendors transition in meaningful numbers to a new technology as a replacement for a legacy technology, it is another important indicator of a sustainable trend. Vendors typically don’t take on technology changes on a whim. There are too many interdependencies and risks associated with undermining the product experience.
Follow the Demand
You may note that we have two demand-side indicators and no supply-side indicators. Lack of supply can constrain a market’s development, but oversupply is a characteristic of every technology hype cycle. This is true for both the justified and unjustified hype trains. As a result, I look primarily at demand when I want to separate the overhyped from the legitimate.
Large language models are doing well based on demand-side metrics. First, we have the statistic that ChatGPT vaulted to 100 million monthly active users in less than eight weeks. On a smaller scale, you have companies like Jasper AI that went from close to zero business users to over 100,000 in about 18 months. GitHub Copilot has more than a million users. The GPT-3 enabled legal contract writing assistant Spellbook has a waiting list of hundreds of law firms.
We also see a lot of technology companies adopting LLM capabilities to build entirely new products or add features to existing products. Below are five product categories that are showing signs of rapid LLM adoption and likely represent a permanent change to user behavior and expectations.
1. Coding
AI-coding assistants were the biggest LLM story of 2022 before ChatGPT came along, and it was arguably a more important development within the year. GitHub says Copilot developers are an average of 55% more productive after adopting the code completion tools. And, GitHub is not alone. Amazon has introduced CodeWhisperer through AWS; there is also the open source GPT-code-clippy, and Google ML (though still an internal tool for now).
The true productivity gains depend on the type of code tasks assigned to developers. However, the biggest gains today are for code tasks that arise frequently, and development teams tell me the impact is substantial. Software developers are costly resources, and their productivity is a key constraint on new feature delivery cycles. Any significant improvement in developer productivity will have far-reaching impacts.
2. Copywriting
Did you know that Grammarly said in 2020 it has 30 million daily active users? You might think of that product as priming the market for AI writing assistants based on LLMs.
We have already mentioned that Jaspar AI had more than 100,000 users at the end of 2022. That reflected a 45% rise in the last three months of 2022. The company also acquired Outwrite at the end of the year, adding over 1 million users of its Chrome extension. AI21 Labs claims more than two million downloads of its Chrome extension and a steep rise in users of its new web app client. There are many more AI writing assistants ranging from Copy.ai to Rytr, and Writesonic. Then there are the 100 million ChatGPT users and millions more using GPT-3 powered features in Canva, Picsart, Microsoft Designers, and so on. Writing will never be the same.
3. Search
This has been bubbling up for some time. ChatGPT highlighted how conversational search could provide a different, and sometimes better, search experience than Google’s ten blue links. Perplexity.ai, You.com, and Bing all offer this new, chat-based search without many or most of the ChatGPT hallucination problems. Google announced its own search assistant named Bard in February. It is not yet available but is expected in the first half of this year. Even Brave’s browser has added custom LLM functionality to its search engine.
Traditional search will not disappear overnight. However, LLM-based search results appear destined to become standard features in modern browsers circa 2023. This is increasing search costs, but search providers are betting the improved user experience will become popular and could be a way to break Google’s stranglehold on the market.
4. Web browsing
The transformation of web browsing by LLMs was less expected, and adoption is likely to take longer than for search. However, Bing and Opera have both introduced website summarization features, and Brave seems likely to follow this lead soon. Key holdouts so far include Chrome and Safari, which means the two leading browsers by market share could wind up suppressing adoption simply through a lack of consumer awareness. Granted, Google has everything it needs to provide this feature through Chrome anytime it wants to push an update.
Still, the integration of LLM-enabled conversational search and website summarization seems inevitable. These features are closely related and provide a valuable benefit for users. We just don’t expect pervasive use will be achieved nearly as quickly as for the other use cases listed here.
5. Conversational bots
Nearly every conversational bot-building platform has added some LLM-based features since December 2022. In most cases, these features augment existing NLU-based models for chat interactions or provide new design capabilities. And they appear to have immediate benefits.
The key concerns around LLMs and conversational bots include hallucinations (i.e. the propensity to produce incorrect information) and the risk of producing off-brand content (i.e. saying things that might reflect poorly on the chatbot host). We expect to see these to be important areas of development in order for LLM-augmented chatbots to become commonplace, but their absence is unlikely to stop this trend.
What do you think? Do you agree or disagree with the entire list? Do you think some of these products are being disrupted, but others here are just overhyped? Why? I look forward to hearing your thoughts in the comments below or on LinkedIn.
Just reading about You.com in your post and decided to see for myself. I asked when daylight savings time begins and was told authoritatively it begins March 14, 2023. That’s actually next Tuesday, while the real DST starts on March 12. Nothing to worry about here….