While I'm not ready to talk about what I am working on just yet. I did want to talk about HOW I have been going back and forth with AI to soundboard, and figure out the best path forward. If you don't like AI then you're free to scroll away, but if you're mixed or do like it, then keep reading.

The project started out as a collection of youtube videos on the subject, with a few articles in as bonus, and then I tossed them all into Google NotebookLM. I might add that its really hit or miss, sometimes it really works well and its shocking, and other times it doesn't even have the context memory of the previous chat message. Super weird, and so I didn't really USE that tool much, but I wanted to get a bearing on what other people have done on a similar project.

I would argue what I am doing is more deeply useful, and not surface level like some of those videos. However I'm still new to it all, and I don't have coding skills. I lack even basic ai agent, or mcp type knowledge.

However what is cool about using Google ai is that your notebooks are now in Gemini, so you can create numerous chats with that notebook. Giving your Gemini chat the proper context it needs. Although to be frank Claude did most of the heavy lifting in this project, and I kept running out of tokens.

I even did the Google Antigravity grift, where you sign up for the pro plan temporarily, and get free Claude credits, but it resets every week and I wasn't going to wait that long so I cancelled.

At one point I ran out of Gemini even (which is harder to do) AND Claude, so reluctantly I went over to ChatGPT (I also tried Perplexity again).

Meshub.ai – Compare Multiple AI Models
Compare and use multiple AI models in one place. Join the Meshub.ai waitlist.

Was really great when I used it in the past, however I haven't been able to log in lately.

MultipleChat: All AIs in One Workspace
MultipleChat brings AI comparison, collaboration, web research, files, documents, presentations, and workflows into one workspace.
MultipleChatMultipleChat Team

Fun fact I applied to them, and got no response. Not even an email declining my application.

Each AI is its own Thinker

  • Gemini has the context window and access to YouTube (sometimes works with YT videos sometimes doesn't). I've been able to keep my chat going in Gemini for weeks with no issues (except in notebookLM based chats), whereas Claude will use up your entire day's tokens in one chat if it gets too long.
  • Everyone knows Claude for being better at coding, but I honestly think it has a better THINKING process too. It seems to listen better as well when it comes to directions, whereas Gemini I sometimes get mad at for forgetting the directions. Claude is also really good at big picture stuff too.
  • ChatGPT is more of a middle ground, just a generalist across the board, and really in my opinion doesn't keep up with the others.
  • Perplexity is just a wrapper, but I forgot that it is there. If I had pro then I might have been able to switch between Claude and Gemini models in the same context window, which would have been really helpful.
  • I forgot that Bing exists, CoPilot, and honestly given that the other 3 exist I probably wouldn't use it. However I might throw my project in there, and just see what it says. Get a different perspective!

This is why projects like LLM Council, Meshub, or Multiple.chat, are really interesting as they combine your options into one.

Since I can't afford to pay for the pro plan for even one model. I tried with Claude, but I was upset that it didn't reset my timeout window (that only timed out because the chat was too long so I literally went to the chat, asked one question, it didn't even respond it just said "credits are up"). Thus I got a refund and decided to try Antigravity again.

Then I definitely couldn't do pro across all of these models. I was doing my own LLM council the manual way, by copying what one model said into the other, and Claude would REPEATEDLY catch mistakes that Gemini had made or lacking critical explanation. Alternatively, Gemini would sometimes come up with ideas that Claude never thought to mention.

This is why it is so important to use more than one model.

I do NOT trust these AI companies

Google is not trustworthy for a million reasons, I mean just for example training its ai on your gmail or google photos is pretty ridiculous.

ChatGPT is a mile worse, and has trained it data illegally for years. Not respecting privacy at all. One might argue that Gemini has better data privacy features that ChatGPT. I even deleted all of my chats, and one of my two accounts on OpenAI.

Claude.... I used to say was a hell of a lot better, but with them purposefully destroying books for scanning them in. Its pretty messed up.

Are AI companies scanning and destroying millions of books, including rare titles?
Snopes reviewed internal documents from Anthropic that show they have bought millions of books to “destructively scan” them to train their AI model.
Snopes.comAnna Rascouët-Paz

Out of the three I would still say that Anthropic is the better company. They even pushed back against the U.S. government when it was illegally trying to pressure the company into letting the military use it's AI for tactical decisions. That earned it some points with me.

My point however is not "which devil is the best", but rather just don't use any of them at all...

There are two meanings here:

  1. Chinese AI companies are doing 10X better. Not necessarily 10x in performance, but I will say models like Kimi/Deepseek/Qwen have repeatedly beaten out US based AI's over and over. Rather they are doing better from a company/economic standpoint. The Chinese market isn't after profitability, which is a major win over OpenAI or Anthropic. They HAVE to be profitable because of the economy the US is.

    Using Chinese based AI's is just a better decision for the average consumer like you or I because it is just more financially logical. Hence why companies across the globe are switching to Deepseek for example. Privacy is a concern regardless if its US or China. Might as well go with the cheaper option.
  2. Just run things local! Speaking of Qwen, or even just basic models on an old laptop or computer. Can you run the latest and greatest?? No, but you could run a quantized model on a decent cpu or gpu, and then get some ai agent work done. You could run a middle of the road model, and get your answers you want, etc. I think the future is LOCAL, more and more computers are coming out with NPU's for AI, and they're getting better.

I don't have a stick in the game

I'm not attached to any particular company or model. I like Claude the most, but if something changed where the company was ruined, or the model went 100x the cost. I would just move on.

I have been using AI since Dec 2023 with ChatGPT, and periodically over the last few years. My allegiance has changed again and again based on what worked.

This post was made today because I wanted you to think about how you were using the AI, and if you were using them to the fullest. Times are only going to change, and presumably get more expensive. Learn how to use AI more fully now before that happens!