How can GPT-4 help with everyday life?

Most talk about GPT-4 is about work: reports, code, marketing. Yet the model inside ChatGPT Plus is just as handy for the ordinary tasks of a household. It is good at three things a busy person needs: sorting messy information, writing clear text on demand, and suggesting options you can choose from. Those three skills cover a surprising share of daily life, from meals to paperwork to who takes out the recycling.

A Beginner's Guide to ChatGPT lists everyday uses such as planning appointments, organizing notes, brainstorming, budgeting and practising a language. What has changed since the book came out is how you can ask. In late September, OpenAI started adding image input and voice conversations to ChatGPT for Plus and Enterprise subscribers. Images work on every platform; voice is an opt-in setting in the iPhone and Android apps.

What "multimodal" means for you

A model that is multimodal can take in more than one kind of input, here text, pictures and speech. In practice, that means you can photograph a form, a broken appliance or the inside of your fridge and ask about it, and you can speak your request while your hands are busy. Voice uses Whisper, OpenAI's speech recognition system, to turn your words into text, and a new text-to-speech model reads the answer back in one of five voices. When you leave voice mode, the transcript stays in the chat so you can reread it.

The output is still text you should read with care. OpenAI warns that the vision model can misread what it sees, so check any detail that matters.

Worked example 1: dinner from a fridge photo

OpenAI's own launch post suggests photographing your fridge and pantry to decide what to cook. A good prompt adds the facts the photo can't show.

Here is a photo of my fridge and one of my pantry shelf.
First, list the ingredients you can see, so I can correct anything you misread.
Then suggest three dinners for two people that use mostly these items,
each ready in 30 minutes or less. One of us does not eat pork.
For the dinner I pick, give a shopping list of at most three extra items.

Asking for the ingredient list first is the important step. If the model mistakes a jar of salsa for tomato sauce, you catch it before it builds a recipe on the error.

Worked example 2: a clear appointment letter

Writing to an office, a landlord or a school takes time mainly because you want the tone right. Give GPT-4 the facts and the purpose, and let it handle the wording.

Write a short, polite letter to my dentist's office.
I need to move my appointment on Tuesday, November 28 at 10 a.m.
I can come any weekday morning the following week.
Ask them to confirm the new time by email.
Keep it under 120 words, friendly but direct. Leave [brackets] for my name and phone number.

Notice that the prompt gives every real detail. The model cannot know your dates, so if you leave them out, it will invent placeholders or, worse, plausible dates. Read the letter once before sending it.

Worked example 3: a fair chores schedule

Sharing chores is a sorting problem, and sorting is something language models handle well.

Make a weekly chores schedule for a household of three: two adults and a 12-year-old.
Chores: dishes (daily), laundry (twice a week), vacuuming (weekly),
bathroom (weekly), trash and recycling (Tuesday and Friday), groceries (Saturday).
The 12-year-old should have age-appropriate tasks only, about 15 minutes a day.
Balance the total time between the adults. Show it as a table, one row per day.

Ask for a table and the result is easy to print or paste into a shared note. If one person works late on Wednesdays, say so and ask for a revised version; the model will rebalance it in seconds.

Keeping it safe and accurate

A few habits keep these uses helpful. Don't photograph documents that show account numbers, health records or other people's private information. Treat any fact you did not supply, such as opening hours, prices or a recipe's cooking temperature, as something to verify. And remember that voice chat requires chat history to stay on during the beta, so your spoken conversations are kept like typed ones.

What are the easiest wins?

Start with jobs where you already know the right answer when you see it. A dinner idea, a polite letter or a chores table is easy to judge: you can tell at a glance whether it works, and a mistake costs little. Those tasks build your prompting skills quickly because you get instant feedback.

The second easy win is rewriting. Paste in an email you drafted in a hurry and ask GPT-4 to make it clearer and kinder while keeping it the same length. You stay the author; the model does the polishing.

The third is turning a photo into a list. A whiteboard of family plans, a page of handwritten notes or a cluttered pantry shelf can become tidy text you can edit. Check what it transcribed, then build on it. Once these three feel routine, move on to bigger planning tasks, where you'll bring the same habits: give the facts, ask for a format, and check before you act.