Do you subscribe to one or more AI service?

[scripting]

I subscribe to ChatGPT, $20 per month, and am satisfied that I get plenty of value for what I pay. I like the way they're constantly improving the product, adding the features I want the most. No plans on stopping.

But recently I tried using Claude.ai to solve a programming problem, since I heard it was especially good at that. I was very happy with that, but exhausted the free service before I was done. Thinking of giving them $20 per month too.

I've recently started getting more liberal with the news I subscribe to and the streaming services (Hulu, Apple TV, etc). So why not apply the same rule to AI services?

What services do you use and for what kinds of tasks? Are you thinking of expanding or switching?

[andrewshell]

Currently paying for:

• ChatGPT - $20/month
• Cursor.com - $20/month
• Bolt.new - $20/month (canceled after one month)
• Anthropic Credits - $25 of credits (no subscription)
• OpenAI Credits - $25 of credits (no subscription)

Considering testing out:

• lovable.dev - $20/month

[scripting]

@andrewshell -- wow that's a lot of them!

is this for you personally?

any comments on which ones are good, for what tasks?

[kerguio]

Pay for chatGPT… but also have access to others through Raycast Pro - fairly cost effective way to try several models

[djwhelan]

Paying $20 per month for ChatGPT.

I also took advantage of a couple of deals (still available, I think) for some free access to You Pro and Perplexity:

• https://www.f6s.com/company-deals/you.com/1-year-free-you-pro-membership-15534
• https://www.f6s.com/company-deals/perplexity-ai/6-mo-free-perplexity-ai-pro-15457

I am using ChatGPT a lot more since I started paying two months ago. I have not tried the others here yet, but this might remind me to try.

[andrewshell]

any comments on which ones are good, for what tasks?

I use ChatGPT all the time for lots of different things. I've been using Canvas to code more lately.

Cursor.com is AI built into VS Code, and I've been playing around with it. It needs hand-holding but works pretty well for full applications that are too big for ChatGPT and Canvas.

Bolt.new is cool. I lump it into the app generator category like Loveable.dev and v0. It does a great job of scaffolding the app, and the designs are usually good, by my standards. However, after a bit, especially if the app is complex, it gets confused and, in many cases, starts wiping out or rewriting code I didn't want it to touch. This is usually when I move the project over to Cursor.

I haven't done much with the OpenAI credits yet, but I've found a few uses for the Anthropic credits.

First, bolt.new¹ has an open-source version that you can run locally and hook up with anthropic credits.
Second, I've been playing around with Fabric² which is a command line tool that allows you to turn prompts into terminal commands. It can use many different apis, I just chose to use anthropic.

Footnotes

1. https://github.com/stackblitz/bolt.new ↩

2. https://github.com/danielmiessler/fabric ↩

[scripting]

@andrewshell -- it's interesting you mention OpenAI credits, I have some too, and have come across a way to use them that I think will be pretty interesting.

i have a new version of bingeworthy working, same user interface, fresh start for the database, i also rewrote the database code from scratch because i have so much a better understanding of how SQL works and i have chatgpt as my programming partner.

the idea is this:

1. a new field in each rating record called "comments."
2. it's a text field, where you can enter your comments about the show.
3. you and openai are the only ones who can read them. and of course the operator as a guard against abuse.
4. then i'm going to let openai aggregate the comments, and see where that goes.

i want to keep bingeworthy a small community to learn from it and also of course to find programs that i would love.

thought you'd find this interesting.

[sedouri]

I only have the $20/month subscription for ChatGPT that has become more and more worth it as the model improves, but have tried the free trial for Cursor just to get a feel for it. Being able to ask it (Cursor) questions about a code base has been somewhat useful, but I find its ability to produce usable code to be somewhat limited. It's fine with extremely simple tasks, but anything slightly challenging requires a lot of effort to make it even partially functional. But maybe I'm just asking too much of a probability machine?

Anyway, I find ChatGPT to be very helpful for both brainstorming and rubber duck debugging.

[xicubed]

• chatgpt
• chatgpt api
• githubs copilot (although it just went "free" now but maybe with less requests?/models) (use the copilot plugin in VSCode and IntelliJ)

[ChristopherMJ]

I use Chatgpt API.

[jsavin]

I'm only paying OpenAI at the moment, but using Claude.ai in free mode, and both Mistral Instruct and Llama 3.2 locally on my PowerBook M3 via LM Studio. I use Mistral and/or Llama locally for privacy/security for sensitive queries, and plan to (if I ever find the time) try hacking together some agentic functionality to help with tedious or mundane programming work via locally run inference.

I did a back-of-the-envelope a while back on using OpenAI's API with GPT 4 to do this and the cost was equivalent to hiring a full-time mid-level engineer. If I can do a lot of that work locally for the cost of electricity, I could imagine farming out harder problems to the big, so-called "frontier" models but I haven't thought very seriously about how to decide which problems are "hard" or not.

[FTWynn]

Currently I'm subscribing to both ChatGPT and Gemini and using Claude's free version... though since most of my tasks are more framing/writing oriented I've been debating getting that one instead.

I'll do a lot of writing scaffolding, trying to see if I can put in into persona conversations that are useful, generating images for my DnD campaign that I DM, and I'm most recently experimenting with trying to better target focus my social posts to what people actually want to read.

To that end I also subscribe to Writing with AI and Every, which are both pretty good newsletters.

[scripting]

I forgot to mention that I also have a full Gemini account because I bought a snazzy new Pixel 9 Pro which comes with one year of Gemini included. So far that's just meant that I get pop up dialogs all over the place telling me I can use Gemini with Gmail, Google Drive and whatever else. Honestly it's annoying. I do not want these apps to do my writing for me. Please. That imho is not a valid use of AI. I can write for myself thank you very much.

[FLCoop]

Testing Perplexity @ $20/mo

[FTWynn]

That's a great point re: Gemini. I find that the stuff that is being pushed by the big AI companies is small potatoes... stuff they feel confident the tech can do even if it's not really very valuable to users. They're not comfortable promoting use cases and approaches that will come with uncertainty.

So we get "here's a larger auto-reply email based on the chips we've shown you in the past" instead of "here's how you might be able to rethink your workflow to get more out of the time you spend, but be careful of errors."

[one-khz]

As a subscriber to the major AI assistants - Claude, Gemini, and ChatGPT Pro - I've found each platform offers unique strengths. My Pixel 9 Fold came with a complimentary year of Gemini, which has been particularly impressive with its Deep Research capabilities and screen sharing feature that helps guide me through complex computer tasks.

ChatGPT Pro has proven incredibly valuable, especially when I used it to help revise my property leases. After the AI-assisted rewrite, I simply had my lawyer review the changes rather than having her draft everything from scratch. That single use case more than justified the $200 subscription cost. I've come to view ChatGPT Pro as having a PhD-level assistant available 24/7 - an incredible value when you think about it.

One interesting approach I've discovered is having these AI models analyze and critique each other's outputs, which provides valuable insights. For coding challenges specifically, I've had great success with both Claude and o1 pro. It's fascinating to watch these services continually leapfrog each other with new capabilities and improvements.