What you need to know about GPT: AI applications, limitations and implications

If you’ve browsed the internet or perused the news in the past year, chances are you’ve heard of ChatGPT, a generative artificial intelligence (AI) model. And, if you’ve checked it out yourself, you know it represents a significant technological advance. Understandably, as this category of AI becomes more sophisticated, questions arise about the future of work and job security for roles that require writing skills, like bid and proposal managers. 

Fortunately, I have good news. As it stands today, generative AI won’t be taking over bids and proposals any time soon. But, it could certainly lighten your workload if used thoughtfully. Indeed, it probably won’t be long until someone in your organization asks how this technology can work for bids and proposals. So, it’s important to educate yourself now so you can answer confidently and accurately.   

As with many things, the best way to understand is to explore. In this article, we’ll focus on how generative AI like GPT works as well as current capabilities, important limitations and implications for the future as they relate to bid and proposal managers. 

Important background information about GPT 

Before we jump into what generative AI means for bids and proposals, we need to establish a foundational understanding of the underlying technology. This context will be crucial later when we discuss applications and limitations. 

GPT and generative AI: What does it do and how does it work? 

At their core, GPT and other generative AI tools all work the same way. Naturally, the details are complex and technical. But to summarize, generative AI’s large language models are designed to predict the next word based on the previous words. And they do this recursively to continue to predict each word to form complete sentences. Similar to how the keyboards in our phones suggest next words. 

GPT stands for generative pretraining transformer and falls into the category of generative AI. Generative clarifies the output, pretraining conveys that developers “taught” the AI model using a large set of existing data, and transformer describes the technical architecture the system is built on.  

What makes large language models like GPT special is that they are trained with massive amounts of data using deep learning algorithms and natural language processing. Consequently they get really good at making cohesive and logical sentences — the output is usually well composed, logical and human-like. A model like GPT is self-regressive, meaning not only does it predict the next word, but it also uses its own word. This is what enables GPT to generate complete thoughts, blogs, stories and anything else you may have seen on the news. 

GPT was developed by OpenAI. Initially released in 2018, there have already been several iterations of GPT and as of March 2023 the current version is GPT-4. ChatGPT and many other applications use the previous version, GPT-3, as their foundation. But, competitors to OpenAI’s GPT model emerge regularly. Regardless of the model used, generative AI tools are being put to good use in a lot of industries for a variety of use cases. For example, BioGPT answers biomedical questions, Socratic AI is an education-based model and JasperAI creates content. 

The difference between tools is often the amount of data, type of data and approach they use to train their model. For example, GPT-3 was trained on 175 billion parameters including English Wikipedia, website data from CommonCrawl, thousands of books and even Reddit links. Every day the number of available data sources continues to grow at a rapid pace.

GPT-4 illustrates this perfectly as the system reportedly uses more than 100 trillion data sources. That’s 500 percent more than the previous model. The potential of the latest dataset is not yet known, but it could significantly improve AI’s understanding of language, output quality and expand the scope of its capabilities. 

GPT for bid and proposal managers 

Alright, now we’re through the technical stuff and can talk about what this means for the bids and proposals industry. As with any technology, there are clear benefits as well as some key considerations involved. With that information, we can make some predictions about what the future of bids and proposals might look like given the rapid growth of generative AI tools. 

Potential AI applications 

AI has been a big part of bid and proposal management for as long as business technology, proposal solutions and RFP software have existed. From workflow notifications and progress updates to requirements analysis and real-time dashboard reporting the use of AI is commonplace in any kind of response management.  

As AI advances, the ways we apply it to our work do too. For example, AI is used to automate responses and search Q&A pairs from a user’s content library to suggest the best answers. It is also increasingly helping users with content management by identifying duplicative content and recommending content owners for reviews. This new generation of technology is no different, it’s simply a new way to add value by assisting with tasks. 

As the technology continues to advance, you’ll likely start to see more and more business applications integrating with generative AI. In fact, at least one response management solution, used by bid and proposal practitioners, announced an integration with GPT. As we explore some use cases, keep in mind that generative AI is just one more tool to help you be more efficient, prepare compelling answers and collaborate with your colleagues. 

Expand or summarize responses 

Working with page limit submission requirements can be challenging. However, generative AI can help by performing menial tasks for you. If you feel an answer is too brief, you can ask the tool to elaborate on the response to make it more robust. Conversely, if you’re over your page count simply select a few lengthy responses and request a summary. This approach is much faster than manually expanding or trimming text and allows you to focus on incorporating win-themes and tailoring responses to your audience’s needs. 

Improve SME engagement 

How often are you chasing down your subject matter experts? Why is answering RFP questions so difficult for them? I suspect that many SMEs see the questions that require their input and feel like writing the perfect answer is going to take a lot of time that they don’t have. Additionally, it can be hard to know where to start when you’re staring at a blank screen. With generative AI, a proposal manager could offer AI a few prompts and quickly prepare a first draft answer to send along as a template for the SME to edit. 

Remove passive voice and legalese 

It may seem minor, but writing proposals in active voice conveys your message with more confidence and clarity than passive voice. Similarly, answers that are easy to understand are more helpful to proposal evaluators than those that are full of technical jargon and legalese. Fortunately, you don’t need an English or law degree to fix these issues. AI can review any text and update it in seconds, shifting from passive to active voice and legalese to plain language. 

Kickstart content updates 

Hopefully you’re routinely reviewing your content library and boilerplate RFP responses. During these reviews, do you ever feel like the copy is stale? If so, you could ask AI to paraphrase the text. A fresh take on an old response quickly helps you brainstorm new ways to approach it. Now, you can mix your original text with the best elements of the AI version for an even more impactful answer. 

Improve readability 

It can be hard for readers to get through dense paragraphs. And, occasionally it might cause them to glaze over the section entirely. This is an area where generative AI can lend a hand. Rather than manually reviewing the response and trying to find ways to break up paragraphs, shorten sentences and add bullet points, you can ask AI to improve readability. It will replace long words with common synonyms and reformat your answers so they can be more easily understood. 

Current limitations and considerations 

As incredible and advanced as generative AI is, it’s not perfect. Consequently, it’s crucial to understand the limitations and risks of using free tools that aren’t designed specifically for bids and proposals. 

Down the rabbit hole 

As illustrated by ChatGPT’s rise to prominence, generative AI is very engaging. Part of what makes it so interesting (and addictive) is the ability to create new iterations of a command to see how the tool responds and updates the output. Fun for writing quirky poems, but not so useful when you get sucked into the tool but still need to meet tight deadlines. 

It’s easy to spend half an hour or more adding context to your requests thinking that you’re going to end up with a perfect answer you can copy and paste into your RFP. That level of accuracy and relevance is highly unusual. Ideally, a GPT integration for bid and proposal managers would have predetermined parameters that align with the use cases outlined above. This helps keep you focused on the goal — responding to an RFP effectively and quickly. 

Proprietary information protection 

Running answer prompts for RFPs through an open-access GPT tool like ChatGPT is risky. Your RFP responses likely contain a lot of proprietary information that you (and your legal team) want to protect. Most free GPT tools incorporate user inputs into their data model and use this information to answer similar queries in the future. Essentially, they could leverage your input to answer a similar question from your competitors — potentially disclosing sensitive information about your business, its product, and its customers. 

Fortunately, there are GPT integrations that take this into account and protect your business’s data. These integrations still leverage the huge dataset offered by GPT-3, but they restrict the input data to your organization’s instance. Therefore, the data isn’t returned to the larger dataset and your information is protected. 

Robots can only mimic humans 

While GPT’s outputs are remarkably natural, it’s important to remember that it works on a purely logical basis. In other words, the output GPT creates can only imitate its understanding of human behaviors, emotion and reasoning based on source data. Because it doesn’t truly understand, this occasionally results in misinterpretations and inappropriate results.  

So, no matter how good a response is, humans are still absolutely essential to success. Only bid and proposal managers can create responses that are personal, genuine and persuasive, because they have a true understanding of nuanced concepts like emotion, context, motivation, background, pains and urgency that AI lacks. 

Additionally, free tools like ChatGPT don’t have the benefit of training using your content library. Consequently, they cannot identify or reliably replicate your brand’s messaging, persona and tone to compose an answer that’s unique to your organization. So, content created by GPT may require significant edits. 

Inaccurate or incomplete data sources 

Generative AI is only as good as the data that trains it and much of that data probably isn’t relevant to your RFPs. Additionally, it’s worth noting that the dataset GPT-3 is trained on is from was last updated in 2021. Which doesn’t seem like that long ago for humans, but is ages in terms of available data.

All of this means that tools may return answers that are just not quite right, biased or outright incorrect. For example, one subset of GPT-3’s training was a catalog of more than 11,000 fiction and nonfiction books from unpublished authors. However, within that dataset, the quality of writing is somewhat uncertain. It’s inevitable that the books contain grammatical and factual errors, unrealistic dialogue and outlandish situations — all processed and then used to create output by a system that doesn’t necessarily understand hyperbole or emotion.  

Again, the solution is human input. It’s much faster for users to edit an answer than try over and over for a more accurate one. 

Tricky translations 

Because GPT-3 was trained using 175 billion data sources, it does include some non-English capabilities. Generally, it performs well with romance languages and Germanic languages. However, it becomes much less reliable and accurate when used for translation of Asian and African languages. So, if you use generative AI to translate an RFP response in English to another language, it’s crucial to have a native speaker review it.  

Implications 

In 2021, one percent of all data was AI generated. At that time, Gartner predicted that by 2025, that figure would grow to 10 percent or more. It’s clear that generative AI will be part of the future of business. Indeed, many organizations (including RFPIO) are already leveraging it in various ways to enhance efficiency and value. But what does it all mean for you? It depends. 

While generative AI has exciting potential, there are definitely other AI solutions you should invest in first. Certainly, RFP software has automation features that deliver more overall value to your organization and response processes.  

Even the most mature bid and proposal teams who want to adopt a GPT integration need to understand that it’s only useful in partnership with human judgment. Generative AI might do the heavy lifting when it comes to writing first drafts, but proposal managers must still infuse the responses with relevant win-themes, consistency and insights garnered from the vendor-prospect relationship. 

There’s no doubt that this technology will continue to advance in new and exciting ways. Even if you’re not yet ready to adopt generative AI, now’s the time to start watching and exploring how it could change bid and proposal management in the near future. Here’s how: 

  • Hands-on exploration — Try out ChatGPT with nonproprietary information or explore RFP solutions that integrate with GPT and protect your data. 
  • Understand in context — How would your company use this tool? Is there a budget allocated for automation initiatives? 
  • Learn more about generative AI — There is a ton of information available about how AI can be used for different purposes. But for foundational information about GPT-3, I recommend this course from LinkedIn Learning. 

With thoughtful adoption, generative AI can deliver tremendous value to bid and proposal teams. With the time saved, you could balance your workload, optimize processes, curate your content library, answer more RFPs or expand your skill set to play an even more strategic role in your organization. 

 

Join the Conversation