Yeah, it’s mentioned a little, but I would have expected it to be a lot more popular is all. Maybe people are just using these LLMs for other things than I am.
Has it been asked yet in this forum? Would be curious how people use llms.
I mentioned Anthropic and I’ve used Claude for my experiments with Cursor. If I had to generalize, I think this community would tend to prefer more human thought and time delays between queries in the loop. From what I’ve seen, heavy Claude usage tends to be preferred by people who are seeking for the LLM to do more of the work. Not saying you are like that or that Claude can’t be used selectively to good effect.
I use Perplexity and Grok mostly, going back and forth between them.
Perplexity seems to run a little behind on Grok, with Grok 3 being out of beta now. I don’t know about the versions on the other AI’s. The idea of Perplexity choosing the best AI for the particular prompt is attractive, but I really don’t know if it works. In fact, I don’t have much idea how to evaluate AI’s. Of course if they repeatedly hallucinate or give useless answers, I wouldn’t want to use them. Thus far, I haven’t seen that with Grok and Perplexity.
I use ChatGPT, Deepseek, Perplexity, and Gemini. So far I’m using free versions. Its impact feels worthwhile but hampered by mistakes and hallucinations. It’s been useful for computer programming tasks. I use it as a search alternative on occasion. MacWhisper is great.
I’m hoping to experiment with local options in the near term.
I’m not much of an AI user, don’t really see much value for me, so I don’t pay any of the “big AI” companies and instead go with Raycast as I use it anyway every day and it gives me access to many/most/all of the rapidly evolving models without having to commit to anything.
For actual web searches, I now use Perplexity instead of Google and couldn’t be happier.
1 Like
We tested most of the major providers for our choice of corporate account and went with Claude. It was easily the best in terms of quality.
However I also have a Perplexity Pro account for web search. Fantastic tool, like getting a little wikipedia page on any topic you like.
I’m curious what line of business your company is in.
Addendum: Just did a regular Google search. Tried AI Mode. Kinda cool and may be useful.
Not for profit healthcare
Curious what in specific you find better over claude or gemini in Cursor.
Is it Junie AI you are talking about/
Pay for Gemini and ChatGPT. I use it mostly for help with coding and math. I will bounce back and forth when I see problems with a “solution” and want to see what “the other guy” thinks. When I get hallucinations, then I will sometimes bring in Claude to see if anyone can get me over a hump. All these systems tend to have problems with math, nonetheless I find them enormously helpful because I really don’t have a human available to me to bounce ideas off of (I am retired. I am married but my partner is not helpful in this realm.) I have not found one to be notably better than others. I use ChatGPT as my primary go to because I like the Mac app that is provided.
For day to day questions, advice on fixing a gutter or how to use some mechanical tool or how to do something elementary in some application that I use rarely, I simply go with ChatGPT. It is easy and it presents answers very clearly and in these domains I do not tend to have problems with hallucinations.
JetBrains AI has been developed using data from real developers rather than being trained on websites like Stackoverflow (which is why others tend to code badly without security measures), and it uses different LLMs, including their own, for different tasks depending on which is best at that task.
It also integrates with their IDEs. Therefore, it also allows you to query any database in DataGrip , Jupyter notebooks in Dataspell, AI powered project management in YouTrack, etc.
I’ve been watching The Talk Show interview from last week’s WWDC and heard Nilay Patel say that people are using LLM’s “100’s of times a day”.
Really? That must be a massive exaggeration, driven by the hype.
Personally, I’ve got installed the chatGPT app in December and Google’s relentless advertising wore me down to point where I gave in and installed Gemini in April…
And now what?..
In 6-7 months I think I’ve used them 10-12 times. Not 10-12 times a day, just 10-12 times across both apps. Despite the all podcasts, the advertising and hubbub around the LLM bubble I’ve not yet found WTF I’m supposed to be using these apps for.
If I want to research content then I’m looking for product reviews or articles from real people who’ve done the work and present the content.
I guess I don’t see the appeal of an instant message type response from an LLM which is 20-80% accurate, and I’ve possibly no way to determine what’s correct without the repeat work myself, i.e. looking for content from real people who’ve done the work, then comparing it to LLM on LSD.
One the keys to any successful product or service is that the benefits should be obvious, to the point where the product almost sells itself. There shouldn’t be the need to conduct your own research to determine this benefits.
To answer the question, maybe I just don’t.
I have Grok because I of my X account, too. I use it a lot and have a slight preference for it over ChatGPT. Any reasons you can share about why you prefer ChatGPT? (I ask because since Apple seems to be going all-in on it, it might be worthwhile for me to subscribe.)
I have a paid Google Workspace account and got Gemini when they rolled it out to all Business Standard and above users. (and raised the price $2/month). So I’ve tried Gemini on my iPhone a few times but I don’t currently need it. I’ve used Google Search almost every day for 20+ years and can still find the info I need with my first search most of the time.
One of the main advantages, IMO, of using Apple devices is we can take advantage of a lot of cross platform/web based programs. And not just its superior hardware.
Hundreds of times per day sounds too high, but Nilay or Joanna Stern also cited a 20 minutes per day metric for OpenAI’s app. That’s still more than your usage but quite believable.
Any reasons you can share about why you prefer ChatGPT? (I ask because since Apple seems to be going all-in on it, it might be worthwhile for me to subscribe.)
Oh, that’s a hard question! Let me give it a shot.
First, it’s worth me pointing out that even though I have Claude and Google Gemini, I almost always turn to ChatGPT. This might be a little circular, but somehow it has become a habit. And that habit is reinforced by the memory feature which means that the more I use it, the more it knows me, the better it’s answers are.
One example, is that I often use this character “→” as a bulletpoint when I write in linkedin. ChatGPT has noticed this and started using it without me needing to prompt it.
Another example - sometimes it greets me with “G’day”, which no one else does, and it reminds me of my dad 
I guess it learnt that when it helped me rewrite a few emails where I started, “G’day” - another memory feature.
Another reason (and this is also a memory feature) is that I can occassionally ask it things like, “Tell me something I’m good at that I might not realise” or “What are my blind spots”.
And then, there’s the killer feature I love - I can have conversations with it while I go for a drive. Sometimes there’s some topic I need a sounding board fore (or a sparring partner). Sometimes we just start chatting and it helps me figure out what’s on my mind (I coach executives and helping them figuring out what’s on their mind is incredibly useful; ChatGPT helps me use it kinda like that). It’s also helpful for digging me out of tricky situations on Tears of the Kingdon).
And, on the mac, it has the built in microphone so I can dictate into it. Claude only has that on the iPhone and ipad, not on the mac.
Some other features, on the mac: I can have it take a screenshot of an app, and ask it questions about the screen shot. It’s a built in feature. Very handy when I’m getting feedback on slides as I change them.
Hope that helps. YMMV.
3 Likes
Below is a summary of my primary current uses of AI. I use the paid version of ChatGPT most frequently, though I also use Gemini and Claude.
I. Writing Suggestions and Editing Support
I use AI as a super editor, not a ghostwriter. It helps refine my writing and explains why its suggestions may improve on my original version. I’ve instructed it to act as a writing tutor by always explaining its suggested edits. While I don’t always accept every recommendation, I consistently learn from the process.
- Offer feedback to improve clarity, grammar, and flow in my drafted articles, essays, and devotionals. Always preserve my tone, sentence structure, style, and vocabulary.
- Identify and point out redundancies, filler, cliches, or unnecessary repetition.
- Help improve transitions.
- Provide input on how to make my writing more concise without changing meaning.
- Flag inconsistent verb tense, pronoun use, or noun agreement for correction.
- Assist in structuring headings (e.g., verb tense) for consistency in document formatting.
- Recommend alternative wording for unclear or awkward phrases.
- Serve as a thesaurus to look up synonyms and antonyms.
- Suggest edits in complex emails I have already composed.
- Ensure consistency in Scripture formatting and placement across a document.
- Suggest executive summary for long, complex reports.
Titles, Headings
- Recommend more compelling, engaging titles for my existing drafts and draft titles. Suggest best SEO practices.
- Propose clearer or more engaging subheadings based on my section content.
- Suggest heading phrasing to fit the transitional flow between sections.
Writing and Workflow Efficiency
- Compare writing environments to minimize reformatting when transferring drafts between applications.
- Outline workflows for drafting on one device and finalizing on another.
- Troubleshoot formatting problems when copying or pasting between writing tools.
- Reformat text into different file types for copying and pasting in other apps.
- Help ensure structural formatting (e.g., headings, indents) is retained across devices.
Task and Project Organization
- Provide templates for index notes, delegated tasks, and reference tracking.
- Assist with designing automation workflows for moving tasks between systems.
- Recommend ways to integrate notes, follow-ups, and actions into a unified workflow comparing different workflows with different app options.
Tech Support
- Answer tech support questions. I’ve instructed AI to always review the most current official support pages for the app(s) in question.
Visual and Media Support
- Propose image concepts that reinforce the message of my article or devotional.
- Recommend appropriate types of images.
- Create images. For example, the image for my Father’s Day article on biblical manhood was AI generated.
Research and Explanation Support
- Define unfamiliar phrases, idioms, or culturally specific expressions.
- Identify Scripture references and provide multiple translations, include the original Hebrew or Greek when requested.
3 Likes
I go back and forth between ChatGPT and Grok for advanced web searches and editing / arranging my writing. Currently testing out paid ChatGPT and I’ve found it satisfactory with a couple reoccurring issues:
TL;DR
- Image creation is difficult to refine.
- OCR requests for existing publications may be sourced elsewhere, rather than the uploaded image.
Verbose verbiage:
Often, the first result I get in image creation is like 90-95% what I need but becomes increasingly difficult to improve as the number of revisions goes up. According to ChatGPT:
When you request an image that’s 95% accurate and ask for small corrections in follow-up generations, the results often stray farther from the original because AI image generation is non-deterministic. This means:
- There is no “memory” of the previous image’s structure — each image is generated from scratch, even if the prompt is nearly identical.
- Small prompt changes can trigger large visual shifts, due to the way the model interprets and re-weights elements in the prompt.
- Visual randomness (noise seeds) vary each time, unless explicitly fixed — so the model can’t truly “tweak” an image; it reimagines it from scratch each time.
My other issue is with OCR. I frequently use ChatGPT to strip lyrics from sheet music, specifically for church hymns. Often, church hymns have multiple versions, with lyrics being slightly or substantially different. I’ve found that ChatGPT will regularly source the lyrics from another version that it found via an internet search. This can happen regardless of my request to use the image only, and to not use external sources. Often when I note that it’s product is incorrect, and that it didn’t use the image, it will apologize, agree with my assessment, ask to try again… and then repeat the same error. I can usually - eventually - get it to work correctly, but then that defeats the purpose of using it for this work: to save me time, and to eliminate human error.