The advent of Large Language Models and ChatGPT have brought in a paradigm shift to the Conversational AI space. Today, enterprises across various industries are considering the utilisation of ChatGPT for diverse objectives. Though the invention has benefitted many, it has also threatened the business for existing companies which were working in the same space.
Launched in 2004, Gupshup is a similar company, a conversational messaging platform, which had to bring something innovative to stay relevant in the market.
In January this year, they launched ‘Auto Bot Builder’—a powerful tool that leverages the GPT-3 Large Language Model (LLM) and finetunes it using proprietary enterprise knowledge base and domain expertise—thereby resulting in a chatbot specialised to an enterprise, unlike ChatGPT, which is a general-purpose chatbot.
AIM Daily XO
Join our editors every weekday evening as they steer you through the most significant news of the day, introduce you to fresh perspectives, and provide unexpected moments of joy
“With ChatGPT launching in November, there was an inflection point in terms of what you can do with LLMs. Our goal was to feed content to Auto Bot Builder and it should be able to create a bot in minutes, and that’s exactly what we have done today” Gaurav Kachhawa, Chief Product Officer, Gupshup.io, told AIM.
Building a Chatbot in a Minute
Auto Bot Builder can quickly integrate data provided by a specific enterprise, whether it’s in the form of a PDF or URL, to construct a bot in under a minute.
Download our Mobile App
“So, what we want to do is create a ChatGPT-like experience, but more on the unique data for the enterprise. What we have done is we have taken the language understanding of the LLMs and trained it, finetuned it to specific enterprise data.”
“This is one aspect of the bot journey it automates. Now, if you want to do more directed workflows, let’s say you want to capture information about your customers then you’ll have to do a little more orchestration,” Kachhawa said.
Further, in terms of accuracy, the chatbots created with the help of Auto Bot Builder are as close to ChatGPT with accuracy levels of more than 90%.
“We leverage the language understanding but the content understanding we bring using the enterprise’s data is not there in public.”
GPT-4 Powered Chatbots
Gupshup, which has access to GPT-4, is already helping enterprises build chatbots powered by the newest language model by OpenAI.
“I think GPT-4 is just yet another iteration for us and we will definitely want to harness its capabilities,” Kachhawa said.
“We’ll have to see it in terms of pricing as well. Obviously, it will be more expensive than GPT. So, for certain use cases, enterprises might want to use richer model; hence, we want to give all the offerings to our customers.”
Kachhawa also revealed that GPT-3 and GPT-4 are not the only models Gupshup is experimenting with. “We are also experimenting with GPT3.5 turbo as well as Anthropic’s Claude, LLMs developed by Google and other open-source LLMs.”
“Since all the models are available, we want to give options to our customers,” he said.
How Much Does it Cost?
Recently, OpenAI has made available APIs for ‘ChatGPT’ and ‘Whisper’ at a price point that is ten times lower than their current models. The ChatGPT API costs USD 0.002 per 1,000 tokens.
“When you have a chat, typically it could be around 1000 words. So, you have that pricing per message, which has a certain number of tokens. Besides, our pricing is obviously based on the selection and choice of what underlying model the customer wants to take.”
“Our job is to educate the customers about the tradeoffs. Certain models are going to give you short responses and work well in certain conditions; whereas, others will give you a lot more longer responses, they can create more generative capabilities,” Kachhawa said.
So, the cost is based on consumption. It will vary for different enterprises depending on tokens, use cases and the model that the customer chooses. Furthermore, there is a platform fee which Gupshup charges.
Another question that arises here however is that with the ChatGPT API available—and at such a lower price point—wouldn’t it make more sense for enterprises to build their own chatbots? In this regard, Kachhawa said that for most of the simpler use cases, enterprises can build their own chatbot with the available API.
“However, the moment you go to a large customer that has a lot of diverse use cases and their own proprietary content, you need to build your own models. And that’s what gupshup does: It actually builds models that are tailored to the company and that require special skills.”
Demand among Indian enterprises
Since the launch, more than 1000 different enterprises have shown active interests in Auto Bot Builder. Gupshup is also in the process of implementing the same for more than 20 enterprises currently.
“We’re already seeing a lot of demand across different industries. I think everybody will start with a smaller volume to learn and understand how it’s helping them drive more sales, automate processes and things like that.”
Further, Kachhawa believes that the adoption will only increase because the cost for building chatbots will keep going down.
“It is going to be extremely accessible to everybody and the reason is because from the time ChatGPT launched to the latest turbo API, the cost has gone down 10x.”
Over the next few months, there will be an increased competition among various LLMs in the market, leading to a potential decrease in costs. Additionally, the emergence of open-source models is expected to make building chatbots more accessible and cost-effective.
“In the coming months, the ability to speak and understand different languages will be available at a fairly reasonable cost and the real value is going to be in the use case, [in that you can work around problems] like are you automating the insurance industry or the medical industry or retail industry and can you do more in domain specific understanding, which is where gupshup can add more value,” Kachhawa concluded.