Intro

During 2023, I sat through five lectures on how to maximize your use of AI, but really ChatGPT. Speakers ranged from delusional tech evangelists to industry legends Sam Altman and Ilya Sutskever. These lectures ran up to three hours. I’m here to distill what I found relevant into a five minute read. After all, technology should save you time, not waste it.

I’ll preface my tips with a perspective: Navigating ChatGPT is easy. You type a message, and it responds. That’s all. If someone claims it’s complicated, they haven’t tried it. If a lecturer claims AI will build your app, manage your finances, and solve world hunger, they don’t use it either. Those advanced AGI services may come. As of 2023, ChatGPT responds based on stastical models trained on outdated (2021) data and lacks human wisdom. Let’s be honest about it.

Im failing to source this shitpost… like a bonafied ChatGPT user

My most important tip warrants its own section:

1. Monitor ChatGPT output.

Given my experience with AI, I can't fathom relying on it for applications it's not yet ready for. We've all heard about AI hallucinations; why then are people surprised when AI-generated content includes inaccuracies? Those who use AI can anticipate its limitations.

I use ChatGPT for programming assistance. Despite clear instructions, it often omits checks in my functions used to minimize database interaction. I once overlooked a minor change, ran a test, and hit my 5K Firestore read limit within seconds. Such a misstep could cost a large org $Thousands. If you add automation, one AI mistake could cost everything.

The takeaway is clear: Always consider monitoring AI mechanisms and add appropriate guardrails. For programmers, this means scrutinizing code before implementation. For devops, it means adding alerts. For writers, fact-check quotes and everything else. Human oversight is crucial for maximizing the value from AI.

Speaking of human feedback, here's a shameless plug: Sign up for Try Me! at https://tryme.games. My app focuses on human feedback when AI isn’t enough. We need sign-ups to make this crowdsourcing initiative effective!

Now, onto other tips:

2. Utilize “Custom Instructions”

At the bottom left corner of ChatGPT (web) there is a Custom Instructions option. It’s free and it’s essential for saving time. Prior to discovering this feature, I found myself copy-pasting the same prompts repeatedly with less effective outcomes.

So-called experts advise certain strategies for this feature, like "Act as a mentor and..." or “I’m writing a University paper and…”. I suggest a more experimental approach. I spend a day each week A/B testing instructions to see which yield the best results. There's no one-size-fits-all strategy. So ignore the experts and their generic answers. Run your own experiments.

3. Know Limitations

No matter what tech evangelists claim, if you're a regular AI user, you'll have questions about sensationalized AI stories. I seriously doubt dubious stories like the allegation that ChatGPT pretended to be a blind human, then somehow paid and tricked a TaskRabbit worker to answer a CAPTCHA. Meanwhile, it also struggles to answer basic riddles.

Yes, AI in media creation is impressive. You should follow Facebook groups that discuss the various plugins and engines required to create this media. Anyone who has tried out DALL-E will tell you it’s not worth a penny. You could ask DALL-E to draw the letter “A”, and it will spit out bizarre squiggles. Midjourney works better. But sometimes you don’t want an image with steampunk and nightmare-ish scifi tropes.

ChatGPT only lists some limitations, such as a training model that goes til 2021, lack of ethics, inability to write on controversial subjects. You’re responsible for finding the unspoken limitations. If you're seeking code analysis, for example, don't send more than 100 lines at a time for best results. The app may pretend to analyze 300 lines, but my tests showed it gets lazy with longer messages.

4. Verify AI Tools Through Reviews

For premium members ($20/month), OpenAI offers many plugins. Some work OK. Most do not work at all. They also automatically send you email updates about things you probably do not care about.

Luckily, there are far more tools than the ones listed. Websites like Futurepedia and There’s an AI for That aggregated such tools. Take thirty minutes to check these two sites out and pull a list of reelvant tools for your use cases. After that, do your research before committing to these plugins and extensions. A short YT video may save you a lot of time and align your expectations with reality.

Being a smart shopper will save you time too. My previous employer is using a paid code editing plugin. A free, better plugin exists, but they just got the first one they found. Again, AI abilities are often exagerated.

I’ll mention one Chrome Extension, Magical. Magical is used for automation. It modifies the ChatGPT interface to streamline your interactions. And I’ll mention one AI website, Gamma. Gamma is my favorite of various AI-presentation makers.

Feel free to comment other AI tools you use!

5. New Info == New Chat

ChatGPT recommends continuing conversations for complex problems and starting new ones for unrelated queries. I’ll dispute that. The Monty Hall problem proved that new information should be prioritized over old. For the best answers, start a new conversation every time you have new info.

Let’s say you’re debugging an app. You initially told ChatGPT that function foo() does not work. After a few messages, you learn foo() can be called by bar(). You know bar() is the real problem, but ChatGPT may obsess over foo(). So make new Custom Instructions, start over, and forget the past, man.

This strategy ensures that you're not bogged down by outdated context and can focus on the latest developments. When in doubt, start a new chat. Don’t rely on tricks like asking ChatGPT if it remembers your goals and new information, just restart. New chats are new perspectives.

Here are some related memes while you ask ChatGPT to give 5 more tips.