Chatgpt is useful for all kinds of things, but there are also a lot of disadvantages. That is why I decided not to use it anymore. This is why – and especially – use this AI instead of chatgpt.
Chatgpt – you use it too much
Yes, I am guilty. I used chatgpt to do all kinds of small jobs at work that I didn’t feel like. I also had myself transformed into a muppet, a Pixar character and a Studio Ghibli figure. And yes, I even tried to write whole articles – although the latter never succeeded so well that I actually used texts.
Anyway, I have a lot of chatgpt’d. Now I regret it. And that has several reasons. Some are a bit more obvious, for others you have to dive into the details a little more.
Chatgpt uses way too much energy
We have written about it before, but every question that you send to Chatgpt costs quite a bit of electricity. According to current estimates, an average prompt for text generation (including finding the solution for a question) consumes between 1 and 3 Wh. That may not sound much, but with this a 60 W pear can also burn for 1 to 3 minutes. It gets worse if you have images generated. They can quickly use around 20 WH. That has been flowing through a thin wire for 20 minutes so that it gives light.
Deep Research (or in -depth research, as it is called in Dutch) can potentially use much more energy, although that is very dependent on how long the research takes. Estimates vary a lot, but in the worst case you just let that light bulb burn for an hour. If you ask chatgpt for help again next time, first think about a good prompt, so that your energy consumption remains under control.

Chatgpt also consumes water
Something that many people do not think about is the fact that every chatgpt prompt (or from another, similar Large Language model) also costs clean water. If you decide to make yourself a muppet, then up to 12 liters of clean water are needed to cool the servers. 12 liters! Think about it if there will be no drop of rain in the Netherlands for weeks.
Chatgpt is disastrous for your own creativity
Oh it happened so quickly. You need a creative idea for a gift. You have to come up with a slogan for a new product at work. Or make an invitation for a children’s party. No worries, chatgpt is only a browser tab from you removed and comes to the rescue. And indeed, after a few minutes you already have a usable result.
But be honest, is this really the best result? The gift tips completely miss the Personal Toucheven if you had incorporated it nicely in your prompt. At first sight, those slogans all seem creative, but if you really think about it, they don’t make sense. And wouldn’t the invitation have been much nicer if you yourself – yes, you – had just started working in Pixelmator? Chatgpt is always the easy route, but you are neither completely satisfied nor proud of your own performance. And that’s a shame.
But what should you use?
Do you no longer want to waste unnecessary water and energy through chatgpt use, but you also don’t want to miss the benefits of AI anymore? Keep it a bit smaller by using an LLLM instead of using an LLM. The first ‘l’ stands for local. An example of this is Private LLM. That means that you prompt not go to a data center, but simply be performed locally on your device. That might be just a little tax for your device and its battery, but certainly the latter is recharged in no time and that does not use much energy either. To pick up the comparison with the light bulb: a full iPhone 13 battery can let a 60 W light bulb burn for about 12 minutes. An AI assignment via your phone lets that pear burn at most for a few seconds. Oh yes, and water is not used at all.

Of course, that does not alter the fact that a local AI model deteriorates your creativity just as much. My advice is therefore mainly to really use it at important tasks where creativity is involved.
Apple Intelligence works largely locally
Incidentally, the majority of Apple Intelligence functions also just work locally on your device. As far as I am concerned, that is ultimately the future of AI: locally on your device, and not somewhere in a data center.