Skip to main content
On This Page
SBM blog CTA mobile 1

Drive growth and reduce costs with omnichannel business messaging

On This Page

Exploring OpenAI's GPT-4o and Its Applications in Knowledge-Based Bots

Gpt 4o cover image
May 14, 2024
Jerry Jeon profile picture
Jerry Jeon
Software Engineer
On This Page
SBM blog CTA mobile 1

Drive growth and reduce costs with omnichannel business messaging

On This Page
SBM blog CTA mobile 1

Drive growth and reduce costs with omnichannel business messaging

In the past couple of months, from Claude 3 to Llama 3, a series of new large language models (LLMs) have been released. These models have significantly improved, drawing keen interest in OpenAI's next moves. On May 13, 2024, OpenAI announced the GPT-4o, widely acclaimed as a revolutionary change. This article will examine how GPT-4o differs from previous models and what it means for knowledge-based bots.

Introduction to GPT-4o

The "o" in GPT-4o stands for "Omni," meaning "all" or "entire." GPT-4o can process text, audio, and image inputs and outputs. I have summarized the improvements and features of GPT-4o as revealed in OpenAI's demo video.

Key improvements in GPT-4o

The previous model to GPT-4o was GPT-4-turbo-2024-04-09, released on April 9, 2024. Here are the improvements introduced by OpenAI compared to the earlier models:

  • Twice the speed of the last GPT-4

  • 50% cost reduction

  • Five times higher rate limit compared to GPT-4 Turbo

  • Support for real-time video and audio

New voice interaction

The ChatGPT app provides an interface called Voice mode, which previously required users to input voice and wait for a response. Now, responses are provided almost instantaneously, in real-time. According to OpenAI, the average response time is about 320 milliseconds, similar to human reaction times. Users don't have to wait until GPT finishes speaking; they can interrupt and interact during the response. The system also allows for the expression of emotions and the creation of diverse voices. The demo showcased commands like "speak more dramatically" and "speak in a robotic tone."

Video features

The demo highlighted GPT-4o's ability to assist in real-time with solving mathematical problems written on paper and providing code analysis by observing a computer screen live.

GPT-4o API Release

OpenAI has also announced support for GPT-4o in its API, per the community announcement. The benefits previously mentioned of twice the speed, 50% cost reduction, and a fivefold increase in rate limit are directly applicable. While these improvements benefit all users, they are particularly significant for API users with high usage rates.

Another significant change is the support for audio input. Previously, for voice-based chatbots, it was necessary to use STT (Speech-to-Text) features to send text to the API. Now, audio can be sent directly to the API, making it much more efficient and enabling the transmission of sounds that are hard to convert to text. OpenAI's goal is to support this feature within a few weeks, and I am excited about this highly anticipated addition to the API.

Conclusions and recap of the GPT-4o introduction

GPT-4o represents a significant improvement over previous models in various aspects. Although some features are yet to be released, the functionalities demonstrated in the demo video indicate a revolutionary change. While the rapid introduction of new models seemed to threaten OpenAI's position, this innovation ensures that OpenAI remains at the forefront regarding buzz and functionality.

The audio input and real-time conversation features are particularly impressive. The successful implementation of GPT-4o's voice conversation features owes much to an excellent user interface. Sam Altman mentioned in his blog that the new voice mode is the best interface he has used, emphasizing the importance of interface in AI technology.

The interface is often overlooked when integrating features such as chat or chatbots into services. Although it might seem sufficient to use the AI model's API to deliver messages, constructing a good chat interface requires significant resources. Sendbird knows this and provides excellent chat interfaces for GPT-4o, Llama 3, and Claude 3 integrations.

Application of GPT-4o in Knowledge-Based Bots

LLMs have limitations, such as knowledge cut-off to a specific date and inability to access private information. To overcome these limitations, knowledge-based bots have been developed. Users can ingest specific information into these bots in various formats, such as URLs, PDFs, and CSV files.

In a previous article comparing Claude, GPT, and Llama performances, I explored which LLM, including GPT-4-turbo, was most suitable for structuring knowledge-based bots. That article concluded that GPT-4-turbo exhibited the best performance in terms of accuracy and conciseness.

Since GPT-4o is an advancement over GPT-4-turbo, I expected it would also demonstrate outstanding performance for knowledge-based bots. When I applied the same tests, the results were as expected, showing excellent question-answering capabilities.

Let's take a look at an example. I registered a PDF file of the “2022 Commuting in the USA” report by the U.S. Census Bureau in the Sendbird dashboard and asked various questions.

Notice the high-quality responses. You can check the complete list of questions and answers on my GitHub.

If you're interested in experiencing the results firsthand, creating and testing your own chatbot is an excellent approach. Sendbird offers a simple process to create a custom AI chatbot tailored to your specific knowledge base in 5 quick steps and only a few minutes. 

Speed: GPT-4-turbo vs GPT-4o

The chatbot on the left uses GPT-4 turbo, and the one on the right uses GPT-4o. When asked the same questions simultaneously, there was a noticeable difference in the speed of response generation.

In my previous article comparing various LLM models, I mentioned that GPT-4's weaknesses were its price and speed, but it seems that these have been completely overcome in terms of speed.

The first no-code, custom AI chatbot for web and mobile supporting GPT-4o

Your chatbot interface is critical. Users expect a chat experience similar to WhatsApp and Telegram, even when talking to a chatbot. When building a custom AI chatbot, it's important to consider a chatbot like Sendbird's that can offer both cutting-edge LLMs like GPT-4o and a world-class chat interface

To maximize the functionality of GPT-4o, trendy and useful chat features should include:

  • Message cards to display product images

  • Suggested replies

  • Message status receipts for sent, delivered, and read messages

  • Typing indicators

  • Offline support

Integrate GPT-4o into your website in just minutes!

Sendbird can assist you in building a GPT-4o powered AI chatbot with no code. You can also train your chatbot with your content using URLs and files through the Sendbird AI chatbot dashboard.

Create your first GPT-4o chatbot today!

Ebook Grow background mobile

Take customer relationships to the next level.

Ready for the next level?