Heiko Hotz

An AI model that reviews legal contracts.

Contract review is the process of thoroughly reading a contract—to understand the rights and obligations of an individual or company signing it—and assessing the associated impact. It is widely viewed as one of the most repetitive and tedious jobs junior law firm associates must perform. It is also expensive and an inefficient use of a legal professional's skills. In this project, I demonstrate how AI can perform this task.

A blog post and code repository that enables readers to deploy their text generation application in their AWS account.

This app democratizes access to state-of-the-art NLP technology and significantly reduces the costs of using it.

I created and implemented an "Impact Strategy Bot" based on OpenAI's GPT-3 model. The goal was to develop an NLP model that generates an interesting strategy for addressing a social problem provided by a prompt from the user.

My task was to create a fine-tuned custom GPT-3 model for an NGO to accomplish this goal. To achieve this, I first analyzed, cleaned, and prepared the more than 4,000 records that the customer provided as a potential training dataset. After diving deeper into the data, I identified potential features that could be used for training GPT-3. I prepared the dataset accordingly and also engineered the appropriate prompts for the model to maximize the likelihood of the model creating interesting strategies. I then trained GPT-3 using OpenAI's API. After a few iterations and testing the custom models with the NGO, I successfully created an NLP model that would generate interesting strategies from a short prompt mentioning a social problem. The model will be implemented by the customer to be used by future social entrepreneurs within their platform.

In this blog post, we will have a look at how we can use the open-source GPT-J model for a variety of NLP tasks, including text summarisation, email creation, sentiment analysis, and code generation. We use different parameters that are common for text generation applications and particular prompts (i.e., prompt engineering) for the GPT-J model.

Although large language models (LLMs) like ChatGPT are remarkable, they come with significant constraints, such as knowledge cut-offs and hallucinations. While this may not be an issue for certain tasks (e.g., using them to generate imaginative content like a novel), there are times when customers and organizations require precise information from LLMs. Rather than conducting direct searches using tools like web search engines, they prefer to access this data via a unified interface (e.g., in a conversational style). In such cases, the accuracy of the underlying LLM is critical. This tutorial demonstrates how to overcome these limitations, allowing you to harness the full potential of LLMs.