My AI Comms Playbook: Your Edge.
I'm sharing practical workflows and my prompt insights to transform your communication strategy with AI
Greetings
We are building a niche substack community at The Rundown Studio. We are getting more subscribers each day interested in receiving my insights on AI and comms in emerging markets. My focus is on making corporate comms, newsrooms and creative teams smarter, faster and better w AI. I’m thrilled each time I see a flash that a new member has joined. Thank you.
Especially to our premium subscribers who will start receiving :
One additional weekly newsletter with deeper dive practical insights
Original and tested prompts by the Rundown team
Premium workflow ideas
Early Access to learning tools we are building
Premium Membership Alert
Our rates have increased for paid subs to $15 a month. It helps us create quality content for you. We want to give you the most important AI x comms thinking you will not get anywhere else. We will share much of it still to free subscribers, but give our paying community a bit more juice with practical tools, skilling suggestions and perspectives from me.
If you are a corporation who would like to hire us to advise on AI strategy please get in touch: zain@therundown.studio
Watch Out for Output
We’ve spent quite some time breaking down granular tasks of corporate comms team and have prioritized the prompts we want to perfect. However when we believe we are in a good place we the LLM does not perform at the level we need. We keep going with trial and error.
Takeaway : It’s hard to make it work accurately. So we need to keep iterating repeatedly until it gets us the quality and caliber of what’s needed as a baseline only.
One useful thing: Giving the LLM really good additional information about the task as possible significantly changes the output quality. Eg. I put a profile of our company in a prompt that shaped the result well.
Talk with Thomas
My co founder and I had an exchange about this and we agreed there are three categories of outcomes.
1. It’s just way off, not following structure and making stuff up.
2. It’s following the structure but just making stuff up.
3. It’s following the structure and the content is right but it’s not how I would do it.
Thomas and I agree that the first two require constant prompting, additional info and context improvements. Point 3 is harder as it requires people to get over a mental hurdle that they have lost agency with this tool. As Thomas says “it’s not a person but it’s not quite software.”
Level Up with Lenny
Here are some more advanced techniques I’ve been using for prompting as you work with your LLMs from
Few Shot Prompting. Take a few examples of what you want it to produce. Add it to your prompt. Eg. Press releases. Emails. Reports. So the LLM knows what good looks like.
Decomposition. The idea here is do not ask the LLM to take on a large task. It will struggle. First go step by step. You ask the LLM what the sub ideas or sub problems are. The building blocks it needs to solve or address. Then solve them one by one. And THEN, handle the overall problem using the granular solutions. This is good for product architecture especially.
Self criticism. Ask the LLM to solve a problem or complete a specific task. The. Ask it to critique itself. “Now go and check your own response and offer criticism of yourself.” You can ask it to implement its own it solution…. And you get a free performance boost.
To go deeper , watch this:
If you’re a communications specialist, a creative or a journalist, I’d love to hear from you. Let me know what you’d like me to focus on in the comments.
Thanks.
Zain.