Taking Writing to the Next Level with and GPT-3’s Text-Davinci-003

Team Pepper
Posted on 29/11/224 min read
Taking Writing to the Next Level with and GPT-3’s Text-Davinci-003

OpenAI is a cutting-edge artificial intelligence company that is transforming the way we interact with technology. Their products are changing the landscape of AI, making it more accessible and understandable for everyone.

OpenAI has released several groundbreaking products over the years, including GPT-3 and DALL·E 2.

1. GPT-3 (Generative Pre-trained Transformer 3): GPT-3 is a large-scale language model that uses deep learning to produce human-like text. It can generate text from prompts, complete tasks like question-answering, and generate text in a variety of styles.

2. DALL·E 2 (Dialog-based Language Learning Engine): DALL·E 2 is a natural language processing model that can generate images from text descriptions and vice versa.

Both of these products have been highly acclaimed by the AI community and have helped OpenAI solidify its position as a leading AI research company.

OpenAI just launched its latest text generation model: text-davinci-003

OpenAI released its latest text-generating model this morning, text-davinci-003 and it is a significant improvement over the existing models. Here’s what OpenAI wrote about the new model:

“text-davinci-003 includes the following improvements:

  • It produces higher-quality writing. This will help your applications deliver clearer, more engaging, and more compelling content.
  • It can handle more complex instructions, meaning you can get even more creative with how you make use of its capabilities now.
  • It’s better at longer form content generation, allowing you to take on tasks that would have previously been too difficult to achieve.”

OpenAI has NOT commented on the exact dataset size of this new text-davinci-003 model, but here’s what they mentioned in their email:

“This model builds on top of our previous InstructGPT models and improves on a number of behaviors that we’ve heard are important to you as developers.”

While text-davinci-002 was already very good at performing tasks and creating content based on instruction, we noticed that plain old davinci performed better on creative tasks. Even within Peppertype, we tuned many models to use davinci as a base instead of the newer models.

With the improvements we’re seeing in our early testing rounds, we observed that the AI performs better in understanding the ‘context’ behind a request and then using that to produce better content.

We’re also noticing that longer generations are better and the hallucination problem that has long plagued AI writing technologies is much solved for. We’ll be testing our complex instructions soon to see the performance improvements.

We saw rhyming come up as a use case too! Not sure if it was intended by the Open AI team or a side effect of improving instruction capabilities. This is what @bokibarum posted on Twitter!

And another one

If you’re already using the previous generation models like davinci or text-davinci-002 for your generations, the newer edition might not be a drop-in replacement in your system. The instruction interpretation, while advanced, will need to be tuned for use cases that you’re trying to achieve. The finetuning advantages will also need to be rebuilt and tested using the newer model.

You can join the discussion on HN here: 

All this is great, but why should you care?

We thought the same, and then we put to the test. We ran our existing content types (developed on top of davinci, text-davinci-002, and other language models) and parallelly wrote newer content types to work on text-davinci-003.

Listing down some of our observations with the new text-davinci-003 model:

If you are a user, you are getting an upgrade on your content quality at no additional cost! That means:

  • More engaging, compelling content.
  • Longer results in certain content types.
  • Lesser irrelevant content created.

Here are some examples of before and after text-davinci-003:

Experiment 1: Quora Answer

Input: What is the philosophy behind WeWork?

The goal of creating a Quora answer is to get an informative and accurate answer to the question. I personally prefer it to be a bit detailed – start with setting a bit more context about the question (in this case, WeWork), then a brief answer to the query, and end by elaborating on it.



Experiment 2: Amazon Product Descriptions

Input: Foamily Throw Pillows Insert 18 x 18 Inches – Bed and Couch Decorative Pillow

The product description is one of the most important marketing copies that explains what a product is and why it’s worth purchasing. But there’s a lot to account for while writing a good product description: focus on your ideal buyer, entice them with benefits (not features!), and tell a story.



…you get the gist!

Experiment 3: Blog Introduction

Start your blog introduction by introducing the topic of your post and piquing readers’ interest. Give a brief overview of the content and explain why it’s important or relevant to your readers. You can also give a teaser about what the post will cover and why it’s valuable. Your introduction should be short and sweet, but engaging enough to draw readers in and encourage them to learn more.



We’re still excitedly running more tests on text-davinci-003 and will keep sharing our results. We’re already hard at work on finding areas of improvement among existing content types on, and users should start seeing the benefits soon.
If you’re looking to try an advanced AI writing tool built for a team, do check out