Marketing Matters: The history of ChatGPT and how to use it in media buying

After its full public release in 2022, ChatGPT took over the headlines and gained 5 million users in just the first five days. But what’s behind this tool and how to use it in media buying? Let’s take a look!

From Neural Networks to Text-Generative AI

Developed in the mid-1960s at the Massachusetts Institute of Technology by Professor Joseph Weizenbaum, ELIZA was one of the first-ever chatbots. Named after Eliza Doolittle from Pygmalion, its primary use was a simulated psychotherapist, specifically designed in a mode known as Doctor. Engaging users with therapeutic prompts, ELIZA often responded to input with questions or statements that mirrored what the user had just said. Despite seeming intuitive, ELIZA could not remember past interactions. Each response was generated based solely on the latest input, without context from previous exchanges.

Even though the LSTM (Long Short-Term Memory) text-generative neural network was introduced in 1997, it gained popularity only in 2010. It operated through a system of gates, each with a unique function. Forget Gate determined which parts of the previous state should be thrown away or kept. Input Gate updated the cell state with new information.

Output Gate decided what the next hidden state should be. This neural network even successfully created music compositions and scripts for short films.

Developed almost two decades after the LSTM, the GRU (Gated Recurrent Units) was released in 2014. It is a simplified version of the LSTM, as it utilizes only two main gates: Reset Gate, which determines how much of the past information to discard, and Update Gate, which decides how much of the new information to store.

Merging some of the functions found in LSTMs, GRUs could be trained more quickly due to reduced computational overhead.

Models like LSTM and GRU allowed machines to remember context over longer sequences, making them more effective at translation and basic text generation tasks.

The Rise of ChatGPT

OpenAI unveiled GPT-1 in 2018. It was a significant step in AI-driven language processing. With only 110 million parameters, it may seem rudimentary by contemporary standards. However, it was the pioneer of a series of more sophisticated models that would soon take over the world.

In 2019, GPT-2 was released. Boasting 1.5 billion parameters, which were essentially tiny weights fine-tuned during its training, it was a marvel in generating coherent and contextually relevant text. OpenAI initially hesitated to release the full model due to fears about its potential misuse in generating fake news or malicious content. Instead, they released a series of scaled-down models to demonstrate its prowess and gauge the public’s reaction.

That was also the year OpenAI and Microsoft solidified their collaborative efforts. With a whopping $1 billion investment from Microsoft and collaboration with Microsoft’s Azure, a cloud platform with immense computational capabilities, it propelled the development of Open AI. 

OpenAI unveiled ChatGPT as we know it today in November 2022, and it instantly became the talk of the tech world. The latest version of ChatGPT has a whopping 175 billion parameters. And even though the library of this text-generative AI only dates to 2021, that’s not for long. In March 2023, the company announced the launch of a plugin that grants access to third-party knowledge sources, including the web. It’s available to select premium users and developers who are on the waitlist. 

According to Microsoft, Bing is integrated into ChatGPT as its primary search engine. This collaboration allows the AI to connect to the internet, offering real-time updates on current events. Additionally, this feature enables the chatbot to reference and cite sources for its responses. Microsoft also outlined its plans to incorporate the chatbot in Outlook, PowerPoint, Excel, and Word soon. 

How to Use ChatGPT in Media Buying

ChatGPT can assist in crafting a compelling sales funnel, eliminating the guesswork. Media buyers can receive specific, actionable insights by feeding it comprehensive data about offers, channels, and consumers. Furthermore, while crafting digital assets like landing pages or creating targeted content, the AI’s ability to design, suggest enhancements, and even pinpoint effective keywords can streamline the process. However, always ensure that outputs undergo human review to avoid potential errors. 

On a broader scale, ChatGPT can optimize campaigns, making them more resonant with segmented audiences and resulting in higher conversion rates. Integrating ChatGPT into your media buying strategy ensures a more data-driven, precise, and efficient approach, maximizing returns on advertising spend.

Remember, while the AI is data-savvy unless you’re using the new plugin, the information in the database only dates to September 2021.

As advertisers, developers, and the general public continue to harness the power of ChatGPT, one thing becomes clear: AI-driven text generation is more than just a tool of convenience. It’s a testament to human innovation, our relentless drive to push boundaries, and a glimpse into a future where machines and humans co-create in ways we once only dreamed of. Let’s see what happens next!