Mouse/Screen Coordinates – Simulating Click on ChatGPT Send Button

I understand exactly what you're saying, and I may actually reconsider my stance on "passing on" keystrokes.

But not separate groups for each app I use KM with. "Never gonna give them up..." :joy:

Me either! In this instance, leveraging group availability conditions (Available when a focused window title contains:) would mean creating a new group just for that particular site (chatgpt.com), and just for one particular macro. Sorry if I didn't explain what I meant very well earlier.

Thanks for getting the reference. :joy:

And we're cool. :sunglasses:

Oh no, I've been Rickrolled! But I'm never gonna give up my separate groups for each app either.

2 Likes

Thinking outside of the box, what kind of queries are you using ChatGPT for? would you be able to use Ollama and a model like Llama 3?

Then it would run in a terminal window

bhowells:

Thinking outside of the box, what kind of queries are you using ChatGPT for?

Anything that comes to mind. If there are things it can do that I don't know about, then I guess they're just simple text queries.

would you be able to use Ollama and a model like Llama 3?

Maybe, but I've never heard of these and don't know what they are.

Then it would run in a terminal window

I open Terminal only rarely, not being connected to an office UNIX system anymore. If the advantage in my case would be being able to send my query by hitting return no matter what the window width, I'm now able to do that with the macro I posted. But thanks for the suggestion anyway. What's the advantage or what are the advantages of doing ChatGPT in Terminal for you?

@Roy_McCoy Thanks for starting this thread. Because of you, I just started using it, and it's awesome!

It's really kind of amazing. I've done Keyboard Maestro searches and gotten great results. Same with Javascript, html and css.

I mean, I've only been using it for a day, but so far it's saved me a lot of time. I'm starting to go to it first, before going to Google.

Ollama is an open source locally run LLM (essentially ChatGPT just not quite so good. - https://ollama.com/

If it's just things that aren't super in-depth i've found it to be pretty useful.

Theres a number of different models you can use depening on what you're doing.

Yeah ChatGPT is great, i've been using it for a while now, I was more asking to see if a locally ran LLM would meet his requirements. Ollama has the benefit of being locally ran, so your data is yours.

Have you tried Claude? @alltiagocom turned me onto it and I've found it to be wayyy more reliable than Chat GPT for code stuff. YMMV etc.

1 Like

DanThomas:

@Roy_McCoy Thanks for starting this thread. Because of you, I just started using it, and it's awesome!

Yes, ChatGPT is awesome. I had Google Gemini as my default AI because ChatGPT in my browser wasn't giving me the cursor in the text field or sending my queries when I hit return unless the window was wide enough, but I eventually realized that I was getting more and better answers from ChatGPT so I switched to that and resolved the two problems concerned. The return key > Send button macro is above. Here below is the other one, which puts the cursor in the text field on my 13" MacBook Air. It requires sizing and positioning the browser window, and I don't know if this is possible without doing that.

Web - Go To ChatGPT.kmmacros (3.5 KB)

It's really kind of amazing. I've done Keyboard Maestro searches and gotten great results. Same with Javascript, html and css.

You can also get lousy results. Beware.

I mean, I've only been using it for a day, but so far it's saved me a lot of time.

Yeah. Who needs government disinformation in dozens of Google finds or in long Wikipedia articles when you can get concise, to-the-point government disinformation in a single AI reply?

I'm starting to go to it first, before going to Google.

This is true with me and several people I know, and I understand it's common. Google has gotten worse and worse, so good riddance.

bhowells:

Ollama is an open source locally run LLM (essentially ChatGPT just not quite so good. - https://ollama.com/

If it's just things that aren't super in-depth i've found it to be pretty useful.

Theres a number of different models you can use depen[d]ing on what you're doing.

Thanks, but I'm good for the moment. What I would really like would be something that doesn't try to gaslight me with crap on political matters. But, as with Google and Wikipedia, I can pretty much tell when it's doing that and get the info I want on less "controversial" things.

If you aren't being hobbled by a lack of support in your browser, try

image

...so you aren't so dependent on window sizing and so on.

I hadn't until I was flummoxed by a tricky (for me) CSS layout question. ChatGPT provided answers that I'd either tried before, or simply didn't work. Claude's first go at it was perfect, and it was explained in such a way that it made perfect sense to me. So Claude's definitely in my tool kit now, too.

-rob.

2 Likes

I think it all depends on what you need it for.
For example, I'm no developer, but I understand a few things. ChatGPT is just a nightmare. It's like a little where you have to tell it 50 times "hey, don't do this"... "I already told you NOT to use this", etc.
It can be really frustrating, especially when you then get the usual "I apologize for...", for it to then do the same mistake on the very next response or 2-3 responses down the line.
When it comes to any type of coding stuff, I can say I now rarely ask it anything.

As @noisneil mentioned, I told him about Claude.ai and I can tell you, it's a HUGE difference. Everything is way more clear to understand, the code is 90% of the time good to go with minimal tweaks, etc.

The only 2 things that are a bit annoying to me:
1 - The free version is limited to a certain amount of messages that can go pretty quickly when you are dealing with code, for example, then you have to wait like 3-4 hours for the limit to reset. If you are good enough with the prompts, take EVERY SINGLE THING into consideration and create a very descriptive prompt, the amount of messages you need to get what you need are heavily reduced. I haven't tried the paid version yet, which is also $20, so I can't comment on that.
2 - After a few messages, at least on my computer, I can feel it dragging when it comes to the browser speed. Maybe on faster computers that's not a big issues, and with shorter chats.

To be honest, for the quality I've been getting from it versus ChatGPT, I don't mind dealing with these 2 limitations.

My suggestion is that you just give it a try and see what you think. It's free, so nothing to lose there :wink:

It's so much better, right? Sometimes it almost feels like there's an actual developer behind it. If ChatGPT needs 10 messages to get it right (if at all, sometimes), Claude sometimes needs 2.

I rarely see myself calling Clause any names (yes, I sometimes get to a point of frustration where I have to just type it as if I was typing to a real person, otherwise I explode) haha
ChatGPT on the other hand, it's almost on every chat...
"Hey, don't use this term, because Script Debugger says this and that".
"I apologize. You're right..."
2 messages later, reintroduces the same error, makes changes that I didn't ask for, etc.

I can see lots of room for improvement when it comes to Claude, but overall I've been loving it. ChatGPT is "ok" for trivial stuff or maybe for some final touches here and there, but the hard work I leave it for Claude.

1 Like

This works and I've modifed the macro accordingly, thanks. The advice to add Pause Until actions was helpful, since without the pause I was getting the alert, "Action Failed Invalid result "" in macro "Web - Go to ChatGPT" (while executing Focus Front Browser Field "//TEXTAREA[@id="prompt-textarea"]")." The macro worked and the alert went away after a few seconds, but it was annoying.

Sure, why not, thanks. [...] I now have it, have assigned a keycode to call it, am using it in the free version and have asked for its help with deleting duplicate files and folders on my Mac. I'll see how it does. If my experience is as positive as yours I'll be happy to switch. I've tended to be forgiving about ChatGPT's goofs because they appeared to be par for the AI bot course at the present time. As my grandmother would say, "Blessed is he who expecteth nothing, for he shall not be disappointed."