ChatGPT for Equity Sentiment Analysis

ChatGPT for Equity Sentiment Analysis is a two-day hybrid workshop explores how asset managers can use a large language model (LLM) such as OpenAI’s ChatGPT to compute sentiment ratings on public companies. Delegates will learn about prompt engineering and how to minimise risks and false signals in generating sentiment scores of companies, resulting in similar outcomes as expensive sentiment products. The course also explores variations and improvements on the basic code as well as other uses of LLM for asset management.

Who should attend: fund managers, hedge funds, traders and equity analysts.

Large Language Models (LLMs) and Generative Pre-trained Transformers (GPT)

Introduction to LLMs: ChatGPT and Other LLMs

  • Typical applications of LLMs
  • How LLMs work
  • Using ChatGPT on the web and GPT-3.5 through the API

Building Applications

  • Overview of prompt engineering
  • Building applications such as text generation, summarization, etc.
  • Few-shot learning with GPT-3.5
  • Introduction to embeddings
  • Overview of the OpenAI embeddings API and its usage

Risks Associated with LLMs

  • Understanding main risks with LLMs, such as, hallucinations, bias, consent and security
  • Methods for reducing the risks of hallucinations, such as, retrieval augmentation, prompt engineering, and self-reflection
  • Methods to detect and address hallucinations, including reinforcement learning from human feedback (RLHF) and model-based approaches

Using GPT for Sentiment Analysis

  • Why we chose the OpenAI GPT-3.5 family among the many available LLMs
  • Evaluating GPT-3.5’s native performance
  • Improving performance with embeddings
  • Worked example: computing sentiment ratings on public companies using embeddings
  • Test data: financial sentences with sentiment labels

Deploying GPT and Other Language Models in Production

  • Best practices for deploying GPT in Production
  • Overview of alternative generative models such as Cohere, LLaMA, Alpaca, etc.