As I explore artificial intelligence more deeply, I’m learning a lot about Large Language Models (LLMs). I want to share what I’ve learned so far because I think it can be useful for other product managers (and anyone interested in this topic). Understanding how LLMs work is key for introducing new features using this technology in products.

What Are LLMs? Understanding the Basics

LLMs are advanced AI systems that can understand and generate human language. They are trained on huge amounts of text, which allows them to do things like write essays, answer questions, and even have conversations.

ChatGPT by OpenAI is a popular example of an LLM. Other well-known models include LLaMA by Meta, Claude by Anthropic, and BERT by Google.

How Do LLMs Work? The Simple Explanation

LLMs learn by studying large amounts of text data, finding patterns, and then using these patterns to create new content. The more data they learn from, the better they become. ChatGPT, like other large language models, is trained on an enormous amount of text data, sourced from a wide variety of places. This includes books, articles, websites, and other written content available on the internet. To put it in perspective, the model processes hundreds of billions of words during training, equivalent to reading millions of books.

For comparison, if you were to read one book a day, it would take you thousands of years to read the amount of text that ChatGPT has been trained on. This extensive exposure to data allows the model to identify patterns and make predictions that result in natural-sounding text.

LLMs use a type of neural network called transformers, which are good at understanding context. Transformers work by paying attention to different parts of a sentence simultaneously, rather than reading it word by word in order. This allows them to grasp the meaning of complex sentences and generate text that is coherent and contextually accurate. This makes LLMs very powerful tools.

Why Should Product Managers Care About LLMs?

LLMs are important for product managers because they offer several benefits:

1. Creating New Features: LLMs can help develop advanced features for your product, such as smart chatbots, personalized recommendations, or automated content creation. These features can improve the user experience and add value to your product.
2. Saving Time and Cutting Costs: By automating repetitive tasks, LLMs can make your work more efficient. This allows your team to focus on more important tasks, speeding up the development process and reducing costs.
3. Standing Out from Competitors: Using LLMs can give your product a competitive edge. The ability to integrate AI-driven features can make your product more attractive to users, helping it stand out in the market.

Imaginary Project Examples: How LLMs Can Be Used

Improving Project Management with ProjectGenius

Let’s say you manage a project management tool (similar to Trello) called ProjectGenius. Here’s how you can integrate LLMs into ProjectGenius and the benefits it would bring:

1. Data Collection: Collecting relevant data from past projects, meeting notes, emails, and chat logs can be challenging. Unless you have enough data, you should consider buying data from reliable sources or using existing datasets with similar contexts. This helps ensure you have comprehensive and relevant data.
2. Training the LLM: Training the LLM on collected data involves feeding the model with this information so it learns the specific context of project management. You can use  machine learning platforms with pre-trained models and tools to simplify and accelerate the training process (e.g. Google Cloud AI, IBM Watson or Hugging Face Transformers).
3. Feature Integration. Some of the features you can develop are:
   • Automated Project Updates: The intelligent assistant can scan meeting notes and generate comprehensive project updates, saving managers hours of manual work.
   • Risk Prediction: By analyzing historical data and current project status, the assistant can predict potential risks and suggest mitigation strategies, improving project success rates.
   • Instant Answers: Team members can ask the assistant questions about project guidelines or processes, reducing delays and enhancing team productivity.
4. Testing: Run beta tests with selected users to gather feedback and make necessary adjustments.
5. Launch: Roll out the feature to all users and monitor its performance for continuous improvement.

By integrating an LLM, ProjectGenius could offer more advanced, proactive support, making project management more efficient and effective.

Enhancing E-Commerce with ShopSmart

Now, consider a digital product like an e-commerce platform called ShopSmart. Here’s how LLMs can enhance ShopSmart and the steps needed to implement them:

1. Data Collection: Gather data on customer preferences, purchase history, and browsing behavior. If collecting this data is challenging, consider purchasing data from third-party providers or leveraging existing data sources with similar customer profiles.
2. Training the LLM: Train the LLM on this data to understand shopping patterns and preferences.
3. Feature Integration. Some of the features you can develop are:
   • Personalized Recommendations: The assistant can suggest products based on individual customer preferences and browsing history, increasing sales and customer satisfaction.
   • Automated Customer Service: By handling common queries and providing instant responses, the assistant can improve customer support efficiency.
   • Predictive Inventory Management: By analyzing trends, the assistant can help predict demand, optimizing inventory levels and reducing stockouts.
4. Testing: Conduct A/B testing with a subset of users to fine-tune recommendations and customer interactions.
5. Launch: Implement the feature across the platform and continuously monitor its effectiveness.

Real-World Uses: LLMs in Different Industries

LLMs are already being used in many industries, such as:

• Healthcare: LLMs are already helping doctors diagnose diseases more accurately. For example, IBM’s Watson Health uses AI to analyze patient data, providing useful information that assists doctors in diagnosing complex conditions like cancer.
• Finance: In the finance sector, LLMs are streamlining tasks such as generating financial reports and analyzing market trends. JPMorgan Chase’s COiN (Contract Intelligence) platform uses LLMs to quickly review legal documents and extract important details, saving thousands of hours of manual work. Additionally, companies like Bloomberg use LLMs to analyze vast amounts of financial data and provide real-time updates on market conditions.
• Education: LLMs are enhancing education by creating personalized learning experiences. Duolingo, for example, uses AI to tailor language lessons to individual learners, making the process more engaging and effective. Similarly, Coursera employs LLMs to generate personalized course recommendations and content, helping students find the most relevant learning materials.
• Entertainment: In the entertainment industry, LLMs are being used to create content in innovative ways. Netflix uses AI to generate personalized content recommendations and even to assist in scriptwriting by analyzing audience preferences and trends.

Interesting Numbers: The Impact of LLMs

• Productivity Boost: Businesses that use LLMs can experience a productivity boost of up to 40%. For example, companies like Salesforce and Microsoft have reported significant improvements in efficiency by automating tasks such as data entry and customer interactions.
• Happier Customers: E-commerce platforms that use LLMs to offer personalized product recommendations see about a 20% increase in customer satisfaction. Amazon, for instance, uses AI to suggest products based on browsing history.
• Cost Savings: Automating customer support with LLMs can cut costs by up to 30%. For example, companies like Bank of America and Vodafone are using AI chatbots to handle routine customer inquiries.

Challenges and Things to Consider with LLMs

While LLMs are powerful, they also come with some challenges:

1. Bias: LLMs can reflect biases in the data they are trained on, so it’s important to monitor them.
2. Ethical Issues: Using LLMs raises questions about privacy and job loss.
3. Accuracy: LLMs can sometimes produce incorrect or nonsensical answers.

Conclusion: The Value of Learning About LLMs

For product managers, understanding LLMs can be very helpful. These AI tools can lead to new features, save time, and make products more competitive. However, it’s also important to be aware of the challenges they bring, such as bias and ethical concerns. By carefully considering how to use LLMs, product managers can improve their products and stay ahead in a world where AI is becoming increasingly important.