Is your AI actually getting dumber?
As we sat around the dinner table eating ground beef tacos, my wife was frustrated, explaining to me that her AI fitness trainer kept giving her the same workout plan, despite constantly prompting it to give her a fresh 8 week plan.
Over and over again she made the same request, but she was getting the same crappy answers.
She was using the same tool (chat gpt). Same subscription. Worse answers all of a sudden.
How is that possible?
I asked her “when was the last time you started a new chat?” She looked at me confused, and shrugged her shoulders. “Never I guess.”
Here’s the thing, her AI trainer isn't broken. The conversation had become cluttered
Every AI chat runs on something called a context window, basically the model's short-term memory for that one conversation. Everything you've discussed becomes part of the conversation's context, old drafts, old corrections, old tangents, unless you start fresh. It’s very useful, until it isn't. Feed it five different jobs in one thread and it splits its attention across all five instead of nailing the one in front of you.
Imagine if you had a new curly-haired intern in your business fresh out of college, and in one conversation you told him everything about the business. His role, daily tasks, the organization chart. It would be unreasonable to expect them to perfectly juggle everything you told them while also solving brand-new problems that pop up.
The same is true with AI, though it is smart, sometimes you can overload too much information in one chat.
So if your chat starts sounding dumber, you have 2 options.
Start fresh chats often- if your AI tool supports memory and you have it enabled, it'll still remember helpful long-term preferences even when you start a new chat.
For repeated work that has the same context, in this case an AI trainer, it’s best to use a Project. This is where you store your fitness goals, experience level, what equipment you have access to, likes and dislikes. Projects let you save the information that should always apply, so every new chat starts with the right context instead of relying on one giant conversation.
Here’s one more simple rule. One chat, one job.
Start a new chat when you switch tasks, not just topics. Drafting a sales email and reviewing a contract in the same thread is two jobs wearing one hat. Start a new chat when the answers get generic or repetitive, that's the model drowning in its own history.
And start a new chat any time you'd have to explain "ignore what I said earlier" to a person. If you'd say it to a human, say it to a fresh thread instead.
Want to learn how to create a Project so AI stores your context in specific folders, so you don’t have to worry about AI forgetting stuff you already told it?
Reply “project” to this email and i’ll send it to you.
I need to make this quick video for my wife anyway 🤣
As promised, value in 1 minute.
Your business AI guy,
Matt
PS: One of my Delegate to Dominate students calls this "clearing the workbench" before starting anything new with his AI team. Same idea, it just took him a program to notice it.