Monday 8 December 2025
By Swikblog News Desk
Most days, people open ChatGPT, type one short question, skim the answer, and close the tab. But OpenAI staff say the model can do more — if you change how you use it.
In a recent episode of the OpenAI Podcast, research lead Christina Kim and product manager Laurentia Romaniuk shared six practical tips for getting better results. I tested each of them on real tasks — from understanding robotics concepts to planning workouts — to see what actually changed.
Here’s what happened, and how you can apply the same ideas in your own chats.
1. Ask harder questions so the model “decides to think”
Kim’s first suggestion sounds almost backwards: if you want smarter answers, stop asking only easy questions. Harder, more layered prompts give the model room to reason instead of just summarising the obvious.
To test this, I tried two versions of the same topic:
- Simple: “What is embodied intelligence?”
- Harder: “Explain how a robot could combine vision, audio and touch in real time to navigate a crowded store without hitting anyone.”
The simple question gave me a neat definition — helpful, but forgettable. The harder prompt produced a structured explanation with sensor fusion, timing constraints, and failure modes. It felt less like a dictionary and more like a mini-lecture.
How to copy this: Instead of “What is X?”, try “How would X work in a real situation?” or “Explain X to me as if I need to make a decision based on it.” You’re signalling to ChatGPT that you want depth, not just a headline.
2. Tell ChatGPT who to be before you ask anything
Romaniuk described how simply setting a role — “you are a frontier researcher”, “you are a clinical coach”, “you are a product manager” — can unlock more advanced or better targeted answers.
I tried this with something trivial: my coffee habits. My base question was:
“Why might someone prefer cappuccinos to lattes?”
Then I added a role: “You are a barista who studies coffee the way sommeliers study wine.” Suddenly the answer turned into a thoughtful breakdown of texture, foam structure and flavour balance. Same topic, very different level of insight.
How to copy this: Start your chat with a one-liner about who the model is and who you are. For example:
- “You are a senior software architect. I’m an intermediate developer.”
- “You are a career coach who specialises in marketing professionals.”
- “You are a patient language tutor for absolute beginners in Korean.”
Those extra 10 words can change the tone, structure and depth of everything that follows.
3. Audit ChatGPT’s memory so it learns the right things about you
One of the most powerful — and underrated — features in ChatGPT is memory. Over time, it can remember details you choose to share, like your job, interests or ongoing projects, and use them to personalise responses.
But that only works if you stay in control. OpenAI’s team recommend regularly reviewing what the model remembers and removing details you don’t want it to hold onto long term. In practice, this looks like opening the memory settings, pruning old or irrelevant notes, and deciding what is actually useful for future chats.
After a clean-up, my own experience changed noticeably. ChatGPT “knew” I’m a writer who cares about AI safety, interviews founders, and trains for fitness races. That meant I got fewer generic explanations, and more answers framed in ways I could apply directly to my work or training.
How to copy this:
- Go into settings and review memory every once in a while.
- Delete throwaway details that don’t need to live forever.
- Actively tell ChatGPT durable facts about you: what you do, what you’re learning, and your preferred style.
You’re effectively training your own “personal version” of the assistant inside the same product.
4. Ask ChatGPT to improve your prompts for you
Kim also suggested a very simple hack: if you’re not sure how to ask a question, ask ChatGPT to design the questions first.
When I needed to understand free-electron lasers — the kind of high-energy light sources used in advanced manufacturing — I didn’t really know where to start. So instead of bluffing, I wrote:
“I’m new to this topic. Suggest 10 high-leverage questions I should ask you to understand how free-electron lasers work and why they matter.”
The model replied with a layered question list, from basic principles all the way to industry applications. I could then copy-paste those questions back into the chat one by one. It felt less like guessing at prompts and more like following a guided syllabus.
How to copy this: Use phrases like:
- “Give me the questions I should be asking about this.”
- “Turn this vague goal into a structured prompt.”
- “Rewrite my prompt to be clearer and more specific, then answer it.”
You’re outsourcing prompt engineering to the model itself.
5. Switch personality modes to see the same idea from new angles
ChatGPT now includes different personality modes, from more neutral to more playful or “nerdy.” Romaniuk says she switches modes frequently to understand how each one feels and what kinds of answers they unlock.
In my own tests, asking for a “nerd mode” explanation produced highly detailed, exploratory answers with lots of side notes and caveats. A request for a more “cynical” tone, on the other hand, gave me a sarcastic takedown of buzzwords like “embodied intelligence” that still managed to teach me the core idea.
How to copy this: If you’re stuck, don’t just re-ask the same question. Ask ChatGPT to:
- “Explain this in nerd mode with full technical detail.”
- “Explain this like a friendly sceptic who doesn’t buy hype.”
- “Explain this like you’re talking to a smart 15-year-old.”
Different personalities can shake you out of your usual way of thinking about a topic.
6. “Pressure-test” the model over time on the same hard task
Finally, Romaniuk recommends treating ChatGPT as a moving target: it improves over time, so things that didn’t work months ago may work now. The way to notice that is to keep “pressure-testing” the model on the same challenging problem.
For her, that’s language learning. For you, it might be exam prep, code review, financial modelling or refining business ideas. In my case, I kept returning to Korean grammar. Earlier this year, ChatGPT sometimes missed context or mis-classified formality levels. Recent attempts have been much better at splitting sentences, labelling each grammar pattern, and suggesting realistic example dialogues.
How to copy this: Pick one skill you want to improve, and build a recurring ritual around it:
- Every week, paste in a worksheet, article or problem set.
- Ask ChatGPT to break it apart, explain the hard parts, and quiz you.
- Save the best prompts and reuse them, so you can compare how answers evolve over time.
If you want a deeper dive into how OpenAI thinks about model behaviour, you can read the original news coverage on Business Insider or listen to the full discussion on the OpenAI Podcast.
Turning insider tips into your everyday workflow
Individually, each of these tips feels small: add a role, ask a harder question, tidy up memory. Together, they fundamentally change the experience. Instead of a one-shot chatbot that answers random questions, ChatGPT starts to feel more like a configurable thinking partner that learns how you work.
If you only remember three steps, make it these:
- Set the scene: Tell ChatGPT who it is, who you are, and what you’re trying to do.
- Ask like a researcher, not a search bar: favour layered, real-world questions over simple definitions.
- Review and repeat: audit memory, save good prompts, and revisit the same hard tasks every few weeks.
ChatGPT hasn’t magically become a new product overnight. But by borrowing a few habits from the people building it, you can make the version you already have feel far more useful.
