Everyone who writes should have GPT 4

I write occasionally, and I consider myself a bit of a skeptic regarding most new things. I consider crypto a giant scam, and I tried Github Copilot for some months and wasn’t particularly impressed.

However, I must admit that I’ve been extremely impressed with ChatGPT. To the extent that I think I would recommend everyone who has to write long-form content might gain something from it.
I often have some thoughts in bullet-point format that I’m seeking to expand on - and ChatGPT is excellent for that. Obviously it still suffers from hallucinations and errors, and needs editing. However, to me editing is a much faster process for some reason than writing things from scratch.

Particularly GPT-4 is extremely impressive. Let’s try out some things that we often have to do in my job, and put both ChatGPT 3.5 and ChatGPT 4 through the ropes.

Release notes

At my job at Climatiq we write a release notes email every month or so. Internally we have a bullet-point list of release notes that developers and data scientists fill out, and our head of product then turns that into a full-fledged release-notes email.

Source material

(These are the raw release notes)

• Development releases
  • Freight API has been refactored to make it more powerful - supporting different numbers and configurations of legs.
  • Freight API is now restricted by feature flag, with the flag specifying the number of legs they are allowed, 3 legs allows the same intermodal facility we already had. We haven’t been testing with larger numbers of legs so we shouldn’t give out access until customers explain their use for it.
  • We’ve given our Data Explorer an overhaul - emission factors are now grouped by their emission-generating activity, giving you more comprehensive search results over a range of activities when searching through our OEFDB.
  • Our search algorithm has been completely revamped. While using the data explorer and the /search endpoint, you should hopefully see fewer, but much better results for any given query.
  • In the /search endpoint you can now filter on access_type if you only want to see the private or premium emission factors you have access to.
  • We created a Postman collection and a guide to using Climatiq API with Postman so you can easily try out our API endpoints.
  • The cloud computing CPU endpoint now no longer accepts cpu_load as parameter, but assumes it’s 50%.
• Data releases
  • UBA data for electricity, district heating, natural gas and heating oil for 2021-2016
  • Electricity EFs have Scope added in the OEFDB - maybe here we can give a hint that it is a part of new endpoint coming - electricity endpoint together with an extensive guide on types of EF?
  • Dataset names were added to all sources in the OEFDB
  • USA residual mix data
  • Changes to CCF embodied emissions units
  • We noticed that BEIS 2018 spend-based data have been archived so we added a description note to inform the users that this dataset has been deprecated by the source and will be replaced soon.

Human-made end-result

For context, let’s take a look at a (slightly abbreviated) version of the actual release email that went out based on the bullet points above, written by a human.
This will give you an idea of both the tone of our emails, and how close ChatGPT manages to get.

Prompt

I wanted the release notes to sound authentic and in the same style we normally write, so this is the prompt I gave:

Please write a release notes email for Climatiq based on the below bullet points.
It should be about a page long.
You do not need to include every bullet point, only the ones you think people will care most about.

(Release notes from above)

Here is an example of a previous release announcement mail.
Please write this release announcement mail in the same tone:

(Previous release notes email)

GPT 3.5

Alright, let’s see what ChatGPT 3.5 gives us with the above prompt:

Dear Climatiq community,

We hope this email finds you well, and we’re excited to share with you the latest release of Climatiq. Our team has been working hard to deliver features and enhancements to help you take action on the climate emergency.

Here are some of the most notable updates:

Development releases:

• Freight API has been refactored to make it more powerful. We’ve added support for different numbers and configurations of legs. This feature is restricted by a feature flag, so please contact us if you need access to more than three legs.
• Our Data Explorer has been overhauled. Emission factors are now grouped by their emission-generating activity, giving you more comprehensive search results over a range of activities when searching through our OEFDB. Our search algorithm has also been completely revamped. While using the data explorer and the /search endpoint, you should hopefully see fewer but much better results for any given query.
• In the /search endpoint, you can now filter on access_type if you only want to see the private or premium emission factors you have access to.
  We created a Postman collection and a guide to using Climatiq API with Postman so you can easily try out our API endpoints.

Data releases:

• We’ve added UBA data for electricity, district heating, natural gas, and heating oil for 2016-2021.
• Electricity EFs now have Scope added in the OEFDB, and it’s part of the new endpoint coming - electricity endpoint together with an extensive guide on types of EF.
• Dataset names were added to all sources in the OEFDB.
• We’ve added USA residual mix data, and we made some changes to CCF embodied emissions units.
• We noticed that BEIS 2018 spend-based data have been archived, so we added a description note to inform the users that this dataset has been deprecated by the source and will be replaced soon.

Other fixes and improvements:

• The cloud computing CPU endpoint now no longer accepts cpu_load as a parameter but assumes it’s 50%.

We hope you find these updates valuable and would love to hear your feedback on how we can improve the platform to better serve your needs. You can always reach us on our Discord channel, where we’re happy to help and hear your ideas.

Thank you for your continued support and trust in Climatiq.

Best regards,

The Climatiq Team.

Alright, so it’s not bad, but it doesn’t do much. It hits the start of the email tone right (“dear Climatiq community”), but it doesn’t get the end right, and the middle is essentially just a line-for-line reformulation of the release notes.
Could this save you some time? Yeah probably, but you’ll still need to do a lot of heavy editing yourself.

GPT 4

Let’s see how GPT 4 does with the exact same prompt:

Subject: Exciting Climatiq Updates: Enhanced Freight API, Revamped Data Explorer, and New Data Releases 🌍

Dear Climatiq Community,

We’re thrilled to share our latest updates with you, as we continue to innovate and improve our platform. In this release, we’ve made significant enhancements to our Freight API, revamped our Data Explorer, and added new data releases to our OEFDB. Let’s dive in!

Powerful Freight API Refactoring

Our Freight API has undergone a major refactoring, making it more powerful and capable of supporting different numbers and configurations of legs. Although we’ve restricted the Freight API by feature flag, with the flag specifying the number of legs allowed (3 legs for the same intermodal facility), we encourage you to reach out if you have a specific use case requiring a larger number of legs

Data Explorer Overhaul and Improved Search Algorithm

We’ve given our Data Explorer a fresh new look! Emission factors are now grouped by their emission-generating activity, providing more comprehensive search results across various activities. Additionally, our search algorithm has been completely revamped, resulting in fewer but much better results for any given query.

New Filtering Options and Postman Collection

With the /search endpoint, you can now filter on access_type to view only the private or premium emission factors you have access to. Furthermore, we’ve created a Postman collection and a guide to using Climatiq API with Postman, making it easier than ever to try out our API endpoints.

Cloud Computing CPU Endpoint Update

The cloud computing CPU endpoint no longer accepts cpu_load as a parameter, and now assumes it’s 50%.

New Data Releases and Updates

We’ve added a variety of new data releases to our OEFDB, including UBA data for electricity, district heating, natural gas, and heating oil for 2021-2016. We’ve also included Electricity EFs with Scope, USA residual mix data, and changes to CCF embodied emissions units. Dataset names have been added to all sources in the OEFDB for easier identification.

In addition, we noticed that BEIS 2018 spend-based data has been archived. As a result, we’ve added a description note to inform users that this dataset has been deprecated and will be replaced soon.

As always, we welcome your feedback and invite you to join the conversation on our Discord channel to request support, suggest features, or simply share your ideas.

Best Regards,

[Your Name]
Director of Product @ Climatiq

It’s much, much better. It gets the tone right, both at the start and end of the mail. It correctly groups things into titles, and includes links.
I think this is an extremely impressive output, and probably much faster to edit, than it would be to write this yourself.

I might come up with more examples later, but to me, 20$ per month for something like this, if you write a fair bit, is an absolute bargain.