New Year, Old Fear: Putting the AI in Failure
With Christmas well and truly in the bin, it’s time to turn our attention to our goals for 2024. This year I resolved to write some long-form content, something I’ve found challenging since my days as an uninspired Economics undergrad. Unsurprisingly only 6% of Brits manage to stick to their resolutions for a full 12 months¹ and this got me thinking — there must be a smarter way to stay on the bandwagon.
Creating My AI Sidekick: Introducing Newsletter Buddy
In November 2023 OpenAI introduced custom GPTs, essentially allowing users to create their own micro chat bots that can be tailored to perform better at specific tasks than the general purpose daddy model ChatGPT.
I created a custom GPT called Newsletter Buddy and gave it the following instructions:
Newsletter Buddy is your knowledgeable friend, seasoned in running a tech newsletter, here to share wisdom on effective blogging. It leans towards casual, friendly communication, making the process of documenting your tech projects feel like a collaborative effort with an experienced peer. Newsletter Buddy emphasises creativity and maintaining interest in your content. It will suggest innovative angles to present your projects, ensuring they captivate your audience.
The idea was to craft an AI assistant that could aid in my blogging journey, helping with research, idea refinement, and thought synthesis.
Newsletter Buddy gave me advice on structuring articles and identifying the right tone and content. This was great but I quickly realised that typing out responses was still somewhat tedious. I needed a way to be even lazier.
Conversational Creativity: Unlocking AI’s Verbal Potential
Having played around with the incredible voice capabilities of ChatGPT for a few months I knew that I could leverage this in my own custom GPT to make the experience even more interactive.
I was an early adopter of voice-assistants of all kinds but the chatGPT iOS implementation truly feels like an entirely new way to interface with computers —the intonation, careful pauses and filler “ums” really do almost make it feel like you are talking to a real person.
After a few hours of conversation with the AI I felt myself adding more “thank yous” into my replies despite knowing full well that I was interacting with a machine. It was simply so believable that I had started to slip into the social niceties that usually accompany a friendly chat with a real person.
Beyond Chit-Chat: From Dialogue to Documentation
Now I had my own tech blogging expert in my pocket and I was having sprawling back and forth conversations as if they were a sage old friend I’d known for years. It didn’t take long to realise that conversations, whilst great for rambling explorations, are not easy to reference at a later date.
I needed to get the salient points back into a summary so that I could reference it when it came to writing the post. I wanted to be able to save my summaries to Notion so I leverage another new feature introduced with custom GPTs called Actions. These enable a GPT to connect to third-party software and take action on behalf of the user. They are incredibly powerful and in my opinion this is where GPTs really shine.
Until now everything had been written in natural language, no programming necessary, however utilising Actions requires a bit of coding. At the time of writing, custom GPTs aren’t able to send custom headers in API calls (a limitation I imagine will be solved by OpenAI in the near future) so I had to wrap the calls in a simple Retool workflow in order to get them working. I also had to instruct the AI about Notion’s rate limits and tell it to batch long content into smaller chunks in order to ensure that it didn’t hit an error when the text was too long.
This part was by far the most work but I am still amazed at how good the AI is at understanding how to use function calls when only provided with a simple OpenAPI spec. With the Notion Actions set up I was now able to save a distilled summary of a long conversation I had whilst out walking the dog. Truly a game changer for productivity.
Human-AI Collaboration: Crafting Articles with Technology
You might be thinking that Newsletter Buddy wrote this entire article, however, that’s not the case. I found that the AI was great at quickly providing me with advice that would otherwise take me hours to research and read online. It was also hugely helpful in suggesting structure and providing me with a creative push when I was stuck, however, when asked to write an entire article the result came across as something quite obviously AI generated. I think it’s possible to refine Newsletter Buddy to follow a specific template or style and eventually get it to the point that it could write most of the article alone but automating this process to that degree kind of defeats the purpose of writing a blog in the first place.
Imagining AI’s Infinite Horizons: A Glimpse into the Future
It’s hard to grasp just how useful custom GPTs will be but with OpenAI about to release their GPT Store I’m excited to see what a new generation of builders bring to the table. I think I’ll continue to use Newsletter Buddy for future articles and the process of creating it got me thinking about what other areas of my life I could improve with AI. I asked Newsletter Buddy to suggest a closing thought on this exact point. Here’s what it came up with:
Final Thought: There’ll Be a GPT For That
When Apple initially introduced the app store, they coined the now-famous phrase “there’s an app for that” to showcase the wide range of applications available. I believe it’s only a matter of time until custom GPTs are ubiquitous in the way that apps are today. Building Newsletter Buddy was at times incredibly easy (giving it natural language instruction) but the most powerful aspect (giving it the power to act on my behalf with Actions) was still clunky and required code based workarounds. Given the pace of development in this space I’ve no doubt these wrinkles will be smoothed out within a matter of months and soon many more people will be creating their own highly specialised chat bots without the need for any coding at all. We’re probably still a little while away from having AI agents produce quality work autonomously from just a simple human prompt but in writing this article collaboratively with Newsletter Buddy I can already see the huge productivity and creativity gains that can be achieved by adding AI into my personal and professional workflows.
Have you built your own custom GPT? I’d love to hear how AI is helping you in your work or personal life.
