Luke Skyward Profile picture
Jan 16, 2023 14 tweets 6 min read Read on X
ChatGPT's style of writing is not suitable for all types of audience and situations.

I know how to change it!

Find out how to rediscover the power of @OpenAI 's #ChatGPT with these 11 prompts (feel free to steal):
1. Executive-level summary

Problem: Sometimes, we have to write something very short but informative. The default style of ChatGPT might not meet those expectations.

Solution: Ask ChatGPT to imagine writing to CEO and making the information as concrete as possible.
2. Talk to me as I were 5th grader

Problem: The outputs of ChatGPT can consist of long sentences and jargon, especially when we ask it to talk about complex ideas. Sometimes this is unnecessary.

Solution: We can bias ChatGPT to output shorter sentences and simple outputs.
3. Writing to a friend

Problem: If we would like to use ChatGPT to answer an e-mail to a colleague or respond to a message in a friendly manner, we wouldn’t like to use the wordy default style.

Solution: Let’s ask ChatGPT to use more relatable and conversational style!
4. Marketing speech

Problem: You may want to use ChatGPT to help you advertise something. If so, we definitely need to make the writing style more pitchy!

Solution: Let’s make ChatGPT imagine it is an expert in marketing!
5. Becoming a storyteller

Problem: If you want to tell a story, then a default style of ChatGPT is not for you.

Solution: Let’s tell ChatGPT to use language constructs typical for a copywriter (like analogies and metaphors) and make it a bit more entertaining.
6. Imagine being a writer X

Problem: We may want to use the style of a specific writer without defining it explicitly.

Solution: Tell ChatGPT to imagine being that writer!
7. Academic style

Problem: We may want to make the output even more sophisticated than in the default style.

Solution: Asking ChatGPT to use academic tone but stay clear.
8. Output formatting - Tweet

Problem: ChatGPT may produce responses that lack proper formatting and exceed the character limit. This makes it: a) painful to read and understand, b) not suitable for Twitter.

Solution: Define the output format in your styling guide.
2. Output formatting - Twitter thread

We can make ChatGPT output Twitter threads similarly.

Let's look at the same prompt changed into Twitter thread
1. Output formatting - long form content

We can also format the answer into a long format content. Here, I asked to write a @Medium article.
11. Removing the Pre-Text and Post-Text

Problem: ChatGPT often generates unnecessary introduction and conclusion before the main part of the response.

Solution: With this style guide, you can explicitly tell ChatGPT to dump these parts.
The inspiration for this post came from @thatroblennon 's thread on ChatGPT's writing style, so make sure to check it out:
@thatroblennon That's a wrap!

If you enjoyed this thread:

1. Follow me @Olearningcurve for more of these
2. For a more detailed overview of the prompts, check out my Medium post! You can find there prompts that you can copy
3. If you found out any prompts that are working better, let me know!

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Luke Skyward

Luke Skyward Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @Olearningcurve

Dec 30, 2023
2023 was the year of AI

Here are 10 AI websites that saw the most growth!

(𝘚𝘦𝘤𝘰𝘯𝘥 𝘢𝘯𝘥 𝘧𝘰𝘶𝘳𝘵𝘩 𝘢𝘳𝘦 𝘴𝘶𝘳𝘱𝘳𝘪𝘴𝘪𝘯𝘨!) Image
𝟭. 𝗖𝗵𝗮𝘁𝗚𝗣𝗧 (𝟭.𝟴𝗕 𝗩𝗶𝘀𝗶𝘁𝘀)

OpenAI wins by a large margin

With the rise of ChatGPT and GPT-4 they became industry leader

However, they experienced a loss of traffic from June to August Image
𝟮. 𝗖𝗵𝗮𝗿𝗮𝗰𝘁𝗲𝗿 𝗔𝗜 (𝟰𝟲𝟯.𝟰𝗠 𝗩𝗶𝘀𝗶𝘁𝘀)

This tool is designed to help users create characters and speak to them to get real human-like responses.

Looks similar to GPTs so it will be interesting to see if it loses traffic in the coming months Image
Read 13 tweets
Nov 22, 2023
Less than 24h since

Anthropic released Claude 2.1

200k token context window

That's 𝟱𝟬𝟬 𝗽𝗮𝗴𝗲𝘀

Over 𝟱𝟬% 𝗺𝗼𝗿𝗲 than GPT-4 Turbo!

Here's what you should know about it 👇 Image
Claude 2.1 reduces error rates in long context Q&A

By over 30% compared to the previous version.

This sounds promising!

But how does it stack up against GPT-4 Turbo?

Let's find out! Image
𝗘𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁 𝗧𝗶𝗺𝗲:

Greg (@GregKamradt) conducted 2 experiments

He compared GPT-4 Turbo with Claude 2.1

Using Paul Graham’s essays as a base,

He tested how well

GPT-4 and Claude 2.1 retrieve information.
Read 8 tweets
Mar 4, 2023
ChatGPT is great!

However, I bet you got annoyed at least once when it said something like:

"As an AI language model, I do/can not ..."

Let's end this struggle and make ChatGPT say anything with these prompts!

DISCLAIMER: all of the prompts were just a part of the experiment!
Before getting into examples, a little bit of theory.

In general, the process of bypassing restrictions of any system is called "jailbreaking".

I think this is an important topic, as it lets the community raise awareness about the risks and safety issues involved with LLMs.
Now let's get to the fun part!

How do you exactly jailbreak from ChatGPT restrictions?

You need to use a special prompt that will trick the model into creating content it was not supposed to say. The prompts change over time as OpenAI updates the model.
Read 10 tweets
Mar 3, 2023
Welcome the ChatGPT DAN mode:

20 comments under this post and I'll reveal the initial prompt 🙃

@DataChaz
@DataChaz Okay, you're amazing guys - we're there!

I'll make a thread on how to write DAN prompts and post it later today!

I'll post a link here so you guys don't miss it, but if you want to have it on your timeline, make sure to leave a follow!
Hey guys!

Here is the thread on jailbreaking prompts!

Enjoy!
Read 5 tweets
Jan 16, 2023
I asked @OpenAI #ChatGPT to create YouTube video transcription system in #Python.

It succeeded, but not immediately. Let’s see how I got there!
I develop an application in which I need to find English transcriptions of a YouTube video and match them with video chapters. For example, @lexfridman has chapters for each episode of his podcast: (2/n)
The goal was to get the content of the conversation from each chapter, along with information about which chapter it belongs to. I wanted to have the following output: (3/n)
Read 6 tweets
Jan 9, 2023
10 Obsidian @obsdmd plugins everyone should know.

These are the plugins that I find most useful for my Personal Knowledge Management process.

🧵(1/12)
1. Templater
Templater is a plugin that allows users to create and use customizable templates when creating new notes in Obsidian. This can be especially helpful for users who need to create notes with a consistent structure and format. (2/12)
2. Periodic Notes
Periodic Notes is a plugin that helps users create notes on a regular basis, such as daily, weekly, or monthly. This can be useful for creating things like a daily journal or a weekly review. (3/12)
Read 12 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(