I use it like I use Google, but like I want to have a conversation with Google.
How long did it take to get that function working?
1 Like
2 hours or so, including all of the chat sessions to test it, and manually editing the notes file in between to test slight variations on the prompt, etc. The majority of the time was spent tweaking the prompt wording to make 4o think about what had been added and then testing/iterating. The code for the API call itself is pretty basic.
1 Like
One thing that I’m appreciating about the Google Gemini LLM AI is that it appears to be much more willing to debate with me.
I just went back and forth for a few rounds with it, and it was willing to hold onto its position. (Not arbitrarily, of course. It provided strong points in support and justification.)
3 Likes
Answering title. For programming. Did light two PowerShell scripts with the help of ChatGPT despite not doing anything in this language before.
3 Likes
Gemini, in its current iteration, does not understand my ideas very well; but Claude really does.
Claude is a more useful scaffold for developing and elaborating my ideas.
I wrote the following idea that I was contemplating today:
Gemini’s response was:
To which I replied:
Given the same original writing, here is Claude’s response:
Claude completely grasped the essence and was able to contextualize it and provide a foundation for further elaboration.
3 Likes
Claude is great! 3.5 Sonnet is currently my go-to LLM for anything other than routine requests, which go to my ChatGPT account as I’ve got unlimited 4o access there. I’m consistently impressed with Claude’s abilities and understanding of the nuances of topics.
Have you, or anyone else, tried the new Deepseek model yet? I’ve heard good things about that one in terms of capability, but the ah… data residency aspect… has me a bit wary.
1 Like
ChatGPT is brilliant. It helped me with PLSQL and Visual basic code when I was stucked. Normally it would take many hours even days to figure out what was wrong but with this, it was done in a matter of minutes. Very useful thing indeed.
1 Like
o1 is good for code, but 4o, in my experience 4o produces a lot of errors… if you like ChatGPT for code, then you will be blown away by some of the others. Claude is fantastic for code. 
1 Like
I am using free ChatGPT and after few or more prompts when I narrow down what I actually want I always got final results which worked and was perfectly satisfied, so no need to change it now, but thanks for advices though.
I am using it not for a whole code, when I only describe what I want, but only when I am stuck with something.
Have you guys checked out the ai studio by google? It’s like having your own tech assistant. https://www.youtube.com/watch?v=3cvczHJSRNs&t=432s
I have seperate ChatGPT chats where I’ve assigned the AI a role - a therapist, a spiritual teacher etc.
and then I bounce ideas off of them, ask for advice, get new perspectives, etc.
The new one I’ve just heard of is NotebookLM. This is a free tool offered by Google, and based on their AI search functionality (currently Google Gemini).
If you go to the NotebookLM website, you can upload any of the books that you own in PDF format to a local NotebookLM instance. Up to 20 PDF’s can be uploaded to one instance and it will only be searchable by you. It is in chat format, so you can then ask it any questions you want, and it will base its replies on the texts that you uploaded to it. It will even give exact quotes without hallucinating. Of course it is not 100% accurate, so everything should be double-checked with the books themselves.
So imagine uploading 10 PDF books on neuroscience, physiology, history, etc. or whatever subject you like, then being able to converse ‘organically’ with all that data.
And then, the other main use I have for AI is creating concept art as a source of inspiration and ideas for my artwork.
3 Likes
I recently subscribed to Kagi search. Yes I now pay for a search engine lol.
But they have an assistant feature which has access to some major LLMs including Claude 3.5 sonnet which I love. It can take a prompt to Claude then also scrape additional web information to give a more informed response. Web access can be turned off too if you just want to hear it from Claude.
Claude is my fav. But I have to be really careful about leading questions. It tends to become too agreeable sometimes and I can end up in my own echo chamber of my own thoughts. Kind of like how scientists can sometimes set out to prove a hypothesis and manipulate the experiment to conform to that vs carrying it out as a way to prove or disprove. It’s hard sometimes not to carry in internal biases and Claude picks up on that sometimes and gives me what I want to hear.
2 Likes
Exactly.
Claude is agreeable. Puts the onus on us to discipline ourselves.
But if you’re able to control for your own bias, Claude is an excellent collaborator and scrapbook/latticework for developing your own ideas.
I think that openness and adaptability to the user’s ideas and logical connections is probably precisely what drives the apparent agreeableness in Claude.
ps. Kagi sounds intriguing
1 Like