ChatGPT, Gemini, Claude
A personal comparison of the three big AI players, and why I ended up with Claude.
Dear friends,
Last days in Brazil… 😭
I was testing the three big players of AI the last couple of months: ChatGPT (OpenAI), Gemini (Google), Claude (Anthropic).
I think we all started with ChatGPT, and I sticked to it for quite a while. But over the time I started to feel odd about it. What I was reading in the news about them did not really convince me to give them my trust. Because basically it is a lot about that. Also it felt odd, that it slowly started to build a global memory of my persona and adjusted the answers according to me.
Because I use Gmail and Gdrive and all the other Google services I started to use Gemini more frequently. Back then I liked the more neutral way of answering. Every chat was new and did not connect to a global memory. I think by now Gemini has memory too, but when I used it, it didn't.
I created a chat for specific topics and added a starter prompt into each one of them. For example: In this chat you are a film-nerd. Or in this chat you are a technical creative. That worked well for a while. Then when the chat became very long, I noticed that it started to forget its initial role. That was quite frustrating, because I thought that I have all my helper-chats set up.
And it has Nano Banana, Gemini's image generation feature, which is one of the best on the market right now. But another thing what truly put me off was that Gemini started to analyse all my files on my Google Drive. All of them. Every image. And that's pretty scary. So I will move a big bunch of my files away from Google in the next weeks.
Then I had a look at Claude, by Anthropic. And at least from what I heard is that they are way more concerned about privacy and ethical questions. So that's good. And they allow users to create projects and add specific instructions to them. So whenever I ask a question in my film-nerd project, I get a good answer.
I now have multiple projects with different instructions set up. I even have a project which writes starter-instructions for new projects for me. (That's an important one) Claude also has memory by now, but it feels more controlled. I know what it remembers and can manage it.
As well Claude is on the frontier of AI-programming. I do use it a lot with Cursor and want to look into Claude Code too.
Just thought I might share this with you.
Much love,
Marius
