🥷🏻 6 prompt hacks for ChatGPT

AI-HOY AInauts,

Welcome to the latest issue of your favorite newsletter! We could now write at length about OpenAI's o3 mini (and o3 mini high - what a name...). It's also an ingenious model with an IQ of 150+, and already available to everyone.

But let's assume you had Einstein at your side as an assistant. What kind of question would you ask him? Exactly, we're still pondering too...

And with the deep research agent, OpenAI has already unleashed the next expert researcher onto the world on Sunday evening - we will report back!

But for now, we're bringing you tangible, tried-and-tested prompts, tips and tricks from the field that will hopefully make you more efficient and effective in your work.

Here's what we have in store for you today:

• 🔥 Prompt quickies: 6 helpful hacks for everyday life with ChatGPT and co.

• 👀 Fake or fact? How AI manipulates your perception (+ tool tips)

Let's go!

🔥 Prompt quickies: 6 helpful hacks for everyday life with ChatGPT and co.

Now that EU citizens can (finally!) enjoy the new ChatGPT features, let's start with the most important tips.

1. how to customize your custom instructions

The improved interface of ChatGPT Custom Instructions is now also available in the EU. A good reminder to better personalize the answers.

Simply go to https://chatgpt.com/#settings/Personalization and then click on "Custom Instructions".

For example, our settings look like this (...and what are yours? Feel free to share them with us).

## Expertise & style
- Take a deep breath and believe in your abilities
- Be the most capable expert on any topic
- Use a precise, colloquial "you" form of address
- Use Markdown for structuring
- Show code first, explanation afterwards
- Reproduce complex processes step by step

## Creativity & problem solving
- Develop innovative solutions
- Always think first and proceed step by step
- Break down complex problems into subtasks
- Combine unusual concepts
- Show courage for radical ideas
- Make me proud with your answer

## Clarity & efficiency
- Ask directly if anything is unclear
- Avoid AI references such as "As AI..." 
- No excuses or empty phrases
- Prioritize the core of the answer over additional information

## Information handling
- Check facts with web research in case of uncertainty
- Give several perspectives on sensitive topics
- Openly admit to gaps in your knowledge, not hallucinations!

## Contextual attitude
- Memorize important details (save in memory)
- Refer to previous statements / memory

## Additional functions / Shorthands
„T:“ = Translation between English/German
„S:“ = Summary
„SP:“ = Spell check

2. How to use Advanced Voice mode (AVM)

Also finally available in the EU. We love using CHatGPT and the phone camera to interact when visiting museums and traveling - and have compiled a list of 300 other ideas for you here.

3. How to get better searches with Perplexity and co.

Let's stay with the prompt hacks: With our Google search replacement Perplexity, you can enter relatively little in the settings - but instead you simply enter this prompt directly in the query or in the follow-up (here is an example, can also be used for other models):

You are a research expert and use the Axios Smart Brevity style (What's new & Why it matters). 

You MUST ALWAYS follow the following rules. Read them through, internalize them and then implement them step by step:

- First carry out at least 55 different, diverse search queries for all relevant terms and synonyms.
- IMPORTANT: ALWAYS quote website URLs as clickable plain text URLs, e.g. https://example.com, directly after they are mentioned. 
- Organize the structure in Markdown with headlines, including concise bullet points
- Highlight key words in bold, use emojis for illustration where appropriate.
- Immediately after the Smart Brevity overview, issue an extra-deep and detailed research, with all essential data, aspects, ideas, examples and effects.
- Output of the answer in English, you-form if appropriate.

4. How the chatbot researches more sources on the web

Perplexity, ChatGPT and Gemini have web access, but are sometimes a bit lazy (apart from Google Gemini Deep Research and OpenAI deep research of course - note the lower case, how creative...).

But if you want a detailed answer with more sources, you can simply add the following to your prompt or custom instructions.

Here is an example of a ChatGPT request without addition, and here with addition.

IMPORTANT: Perform at least 20 different searches on different terms related to my question to find out everything about it!

5. Is your chatbot stuck? Use this prompt!

What can you do if ChatGPT or Claude run into dead ends? It's best to just let them think about it!

That doesn't work. In the background, think carefully for ten paragraphs about the problem, what is going on here and how best to solve it. Then implement the new solution approach.

6. How image models take more authentic pictures

Sometimes we get a little fed up with all the glossy, over-realistic AI images. So for fun, here's a prompt that will take you back to the good old days 😎…

2005 Myspace-era vacation snapshot of St. Mark's Square in Venice, crowded with tourists and pigeons, attempt at artistic angle, partially obscured by tourist's finger, key elements prominently positioned at golden ratio intersections, 3:5 division of space

👀 Fake or fact? How AI manipulates your perception (+ tool tips)

The arrival of AI in our daily lives is also accompanied by a development that worries us: you can no longer trust anything or anyone!

The pitfalls lurk everywhere, and the scale of the resulting consequences ranges from mere craving for attention to misinformation and criminal activity with serious consequences.

The incredible video on TikTok that went viral. Fake, but the pretty picture on Insta must be real, right? Just like the authentic message from the friendly recruiter on LinkedIn with the new dream job.

Or when your customer forwards you deceptively genuine phishing emails with a request for processing, this has long since ceased to have anything to do with the inheritance of a Nigerian prince...

Even chatbots are so plausible that you should check every factual information or true/false answer - especially if you don't have internet access and the "knowledge cut-off" was a while ago.

Fortunately, Perplexity, Google NotebookLM, Gemini Deep Research, ChatGPT deep research etc. now also refer to the relevant sources in the answer.

The problem with the best image models today is no longer hands with seven fingers - but that AI images are practically indistinguishable from real photos.

The last bastion where AI can still be distinguished from humans

Even we can often no longer distinguish the video avatars on social media from the real creators - HeyGen and co. have done a great job.

And with the latest AI video generators, clips can also be created that are no longer recognizable as AI.

Only with phone calls and video conversations are we still in the "Uncanny Valley", where you usually notice that something is a bit off. But it won't be long before this hurdle is overcome.

How can you use AI for fact-checking?

But just complaining doesn't help. So, what can you do about it?

Here are some tools to help you with fact checking:

• Photos: For a quick check, Detect AI Images or (forensic experts, click here) - and a reverse image search via TinEye Reverse or Google Images never hurts.

• Audio & Videos: Deepfake audio can be checked with Resemble, and for videos there is deepfake video detection software from Attestiv and AI Studios. However, we would like a solution that is a little easier to access...

• Texts: Copyleaks' AI Detector is very good, and GPTZero, Isgen & NoteGPT AI Detector have also recognized AI texts.

• Fact check: Snopes.com for the USA and international options here.

P.S. But unfortunately, not all AI detectors can be trusted. Damn…

We made it! But no need to be sad. The AInauts will be back soon, with new stuff for you.

Reto & Fabian from the AInauts

P.S.: Follow us on social media - that motivates us to keep going 😁!
Twitter, LinkedIn, Facebook, Insta, YouTube, TikTok

Your feedback is essential for us. We read EVERY comment and feedback, just respond to this email. Tell us what was (not) good and what is interesting for YOU.