Which AI do you use?

[seanfire]

jmk909er said:

AI is everywhere these days, and there are plenty of options out there. My go-to has mainly been ChatGPT, though I've heard people point out that it can get things wrong at times and tends to agree with you or say whatever you're hoping to hear.

For coding work, Codex has become my go-to tool and I've been leaning on it a lot. There's talk that Claude Code could outperform it, so at some point I might make that my primary dev environment instead.

At the moment, though, the real bottleneck for me is that my token supply just isn't sufficient.

 

[Youarewhen]

Sasquatch Hunter said:

Every day I turn to Chat, and I gave it the nickname Bobo, borrowed from a former colleague whose advice was always spot-on.

My GLP-1 and weight loss journey is something it has tracked alongside me. Each pound shed has been recorded, and each pound put back on as well. The ups and the downs alike.

Designing the hat patches I create is another task I bring to it. It isn't my only resource, but it provides a solid starting point, and for image editing it assists me in Photoshop or Illustrator.

A patch for the group is something I need to make.

Which machine do you use for printing those patches? The one from the tint shop looks fantastic!

 

[BNLFL]

None for me.

 

[latviantower]

clay.clayton said:

When I see 'AI', my mind goes to 'Aromatase Inhibitor'. lol

Before I'd even clicked on the thread, “Aromasin” was already in my head. My E2 went from 130 down to 15 in what felt like no time at all.

 

[latviantower]

clay.clayton said:

When I see 'AI', my mind goes to 'Aromatase Inhibitor'. lol

Before I'd even clicked into the thread, "Aromasin" was already in my head. It took my E2 down from 130 to 15 in what felt like no time at all.

 

[cldfront]

ChatGPT is what I pay for, though I also make use of Claude's free tier. The ceiling on how much an AI can help me seems to be set by how well I can formulate a good question. That's a skill I'm trying to push further.

Hat guy - it's obvious those designs came out of AI. Some folks are going to have a real issue with that, even if fewer than I'd like. I realize I'm just yelling into nothing, but count me in favor of human artists and designers staying employed. Which brings to mind Doc Holiday's line in Tombstone: 'my hypocrisy goes only so far' - so that's also a thing.

 

[FunkyWhales78]

cldfront said:

ChatGPT is what I pay for, though I also make use of Claude's free tier. The ceiling on how much an AI can help me seems to be set by how well I can formulate a good question. That's a skill I'm trying to push further.

Hat guy - it's obvious those designs came out of AI. Some folks are going to have a real issue with that, even if fewer than I'd like. I realize I'm just yelling into nothing, but count me in favor of human artists and designers staying employed. Which brings to mind Doc Holiday's line in Tombstone: 'my hypocrisy goes only so far' - so that's also a thing.

It's fascinating, honestly. There's a common assumption that AI simply hands over answers, yet the outcome hinges heavily on how you frame the question—and, I'd argue even more so, on the questions you ask next.

 

[Sasquatch Hunter]

Youarewhen said:

Sasquatch Hunter said:

Every day I turn to Chat, and I gave it the nickname Bobo, borrowed from a former colleague whose advice was always spot-on.

My GLP-1 and weight loss journey is something it has tracked alongside me. Each pound shed has been recorded, and each pound put back on as well. The ups and the downs alike.

Designing the hat patches I create is another task I bring to it. It isn't my only resource, but it provides a solid starting point, and for image editing it assists me in Photoshop or Illustrator.

A patch for the group is something I need to make.

Which machine do you use for printing those patches? The one from the tint shop looks fantastic!

Thanks — when I need full color, the Eufy Make E1 UV printer is what I turn to. As for the leatherette, those get engraved on an Onelaser XRF Laser.

The UV printer falls into the "at home" category and runs a couple grand, though my reasoning was to start with an entry level machine rather than put several thousand dollars into one right away, so I could be sure I had a handle on operating it.

 

[jmk909er]

I've got something I want to ask everyone...A few of you said you add a precondition to your AI before posing a question, telling it to give accurate answers, not just agree with you, cite sources, and so on. Would any of you be willing to share your pre promps? I'd really like to see them so I can save one and paste it in before I ask AI a question going forward. Thanks

 

[Troglo]

jmk909er said:

I've got something I want to ask everyone...A few of you said you add a precondition to your AI before posing a question, telling it to give accurate answers, not just agree with you, cite sources, and so on. Would any of you be willing to share your pre promps? I'd really like to see them so I can save one and paste it in before I ask AI a question going forward. Thanks

The answer really changes based on what you're doing. Evaluating how an LLM handles language is a very different job from developing an app. On top of that, you have to keep refining your approach. The initial response won't necessarily fit, since the model might have misread something or you might not have provided everything. So you fix it or supply the missing pieces. No matter what the task is, this is the framework I follow:

Context:

What background does the AI require about the situation?

Task

What am I asking it to do, and what method should it use?

Goal:

What is the end result I'm after?

Output:

How should the finished answer be structured? If I have something particular in mind, I can drop in an example too.

After that, inside the prompt I'll include instructions such as: tell me if I'm mistaken, point out any blind spots or gaps, put accuracy first, provide a confidence score so I can gauge how sure you are about what you're stating, and look up/cite sources where necessary.

 

[di_peppil]

I use every one of them. Subscriptions I pay for include Gemini and Claude. For fast tasks I go with Gemini; when an investigation involves lots of data over a long stretch, I turn to Claude Code — it actually retrieves material, parses it, and writes scripts to correlate things. Their deep-research features don't appeal to me much: too much fluff, too little substance.

 

[Foggy-Hollow]

Redz said:

My job is training AI systems. The field of AI work is growing rapidly. Out of all the options, Gemini is what I've used the most. AI isn't disappearing. Either people start using it, or they'll fall behind.

What's the real point here — why bother, and to what end? An 8/6/2026 Wired piece titled “Why Normal People Aren’t using AI Agents” drives home something crucial: AI is a tool. If people don't perceive a need for that tool, they won't reach for it. The average person is after a product, not a tool. What they want is the app, not the coding.

A certain segment of us is built to tinker and adjust, to keep poking at the system until the output satisfies us. The majority, though, just want something that does the job and works without fuss.

Sure, AI is revolutionary. My sense is it hasn't matured enough yet.

rixsternaz said:

Redz said:

My job is training AI systems. The field of AI work is growing rapidly. Out of all the options, Gemini is what I've used the most. AI isn't disappearing. Either people start using it, or they'll fall behind.

Where I work, every Engineer has to complete Gemini training

It seems Microsoft struck some arrangement with my employer — we hold monthly “lunch and learns” about using Copilot, and a few months ago everyone received full licenses. I work in a legacy industry where there isn't much that calls for development. My most successful application so far: taking an engineering write-up, running it against a checklist, and having the gaps flagged.

Using it doesn't bother me, and it genuinely raised the quality of my document. Still, the use cases are thin at the moment.

 

[Redz]

Foggy-Hollow said:

Redz said:

My job is training AI systems. The field of AI work is growing rapidly. Out of all the options, Gemini is what I've used the most. AI isn't disappearing. Either people start using it, or they'll fall behind.

What's the real point here — why bother, and to what end? An 8/6/2026 Wired piece titled “Why Normal People Aren’t using AI Agents” drives home something crucial: AI is a tool. If people don't perceive a need for that tool, they won't reach for it. The average person is after a product, not a tool. What they want is the app, not the coding.

A certain segment of us is built to tinker and adjust, to keep poking at the system until the output satisfies us. The majority, though, just want something that does the job and works without fuss.

Sure, AI is revolutionary. My sense is it hasn't matured enough yet.

rixsternaz said:

Redz said:

My job is training AI systems. The field of AI work is growing rapidly. Out of all the options, Gemini is what I've used the most. AI isn't disappearing. Either people start using it, or they'll fall behind.

Where I work, every Engineer has to complete Gemini training

It seems Microsoft struck some arrangement with my employer — we hold monthly “lunch and learns” about using Copilot, and a few months ago everyone received full licenses. I work in a legacy industry where there isn't much that calls for development. My most successful application so far: taking an engineering write-up, running it against a checklist, and having the gaps flagged.

Using it doesn't bother me, and it genuinely raised the quality of my document. Still, the use cases are thin at the moment.

Teaching people how to use AI isn’t what I do — my work is actually the training of AI itself. And the number of roles in this area is growing rapidly.

 

[swimmer]

jmk909er said:

I've got something I want to ask everyone...A few of you said you add a precondition to your AI before posing a question, telling it to give accurate answers, not just agree with you, cite sources, and so on. Would any of you be willing to share your pre promps? I'd really like to see them so I can save one and paste it in before I ask AI a question going forward. Thanks

I ask Gemini to quit telling me it isn't offering medical advice and that I should consult a medical professional. I also instruct Gemini to stop guessing at my emotions and just answer what I asked. At times, that does the trick.

 

[swimmer]

Grok suits me far more than Gemini does, yet Grok caps how long I can use it.

 

[Troglo]

Redz said:

Foggy-Hollow said:

Redz said:

My job is training AI systems. The field of AI work is growing rapidly. Out of all the options, Gemini is what I've used the most. AI isn't disappearing. Either people start using it, or they'll fall behind.

What's the real point here — why bother, and to what end? An 8/6/2026 Wired piece titled “Why Normal People Aren’t using AI Agents” drives home something crucial: AI is a tool. If people don't perceive a need for that tool, they won't reach for it. The average person is after a product, not a tool. What they want is the app, not the coding.

A certain segment of us is built to tinker and adjust, to keep poking at the system until the output satisfies us. The majority, though, just want something that does the job and works without fuss.

Sure, AI is revolutionary. My sense is it hasn't matured enough yet.

rixsternaz said:

Redz said:

My job is training AI systems. The field of AI work is growing rapidly. Out of all the options, Gemini is what I've used the most. AI isn't disappearing. Either people start using it, or they'll fall behind.

Where I work, every Engineer has to complete Gemini training

It seems Microsoft struck some arrangement with my employer — we hold monthly “lunch and learns” about using Copilot, and a few months ago everyone received full licenses. I work in a legacy industry where there isn't much that calls for development. My most successful application so far: taking an engineering write-up, running it against a checklist, and having the gaps flagged.

Using it doesn't bother me, and it genuinely raised the quality of my document. Still, the use cases are thin at the moment.

Teaching people how to use AI isn’t what I do — my work is actually the training of AI itself. And the number of roles in this area is growing rapidly.

Hey! Pretty similar situation on my end! I lean more toward the LLM QA area. It's certainly an expanding field, though it remains very competitive at the moment given all the SWEs laid off in recent years

 

[Tagget]

My experience with AI has mostly been on the job, and it’s a fairly new thing for me. CoPilot is what we rely on because it slots neatly into the rest of the Microsoft ecosystem, though in my view it remains among the least capable options. When it comes to my own preference, Claude takes the top spot—it has genuinely stood out to me.

 

[lessthanhalf]

What strikes me as fascinating regarding AI in medicine is how heavily the outcome hinges on whoever is typing, the wording they choose, and which details they treat as relevant. Take plain chatgpt: it performs at least on par with a typical doctor, arguably superior, sitting somewhere around a 2nd year specialist trainee, shy of true specialist level — yet that only holds when a doctor is feeding it information, or when the input resembles standard written clinical notes. This matches my own observations. It "knows" considerably more than I do, outperforms my GP, and occasionally beats a specialist, though certainly not consistently. Over roughly a year, the leap from chatgpt 4 to 5.3 has been substantial, and I expect that trajectory to keep going.

Flip the scenario, though, and let an ordinary person or patient supply the data or describe their ailment, and the quality collapses far beneath what an average doctor delivers. Serious errors become possible — for instance failing to flag that you must get to hospital immediately for something virtually no doctor would overlook. In real terms, this makes seeking health guidance from chat bots a genuinely risky move for the general public, since so much rests on the specifics provided and the vocabulary employed. Still, at least it sometimes beats having nothing at all, which could be the case when cost or access barriers stand in the way — particularly in specialist fields where getting seen is hard or painfully slow.

Nobody has offered me a convincing account of why this gap in performance is so enormous. Could it be that solid medical training material exists in medical language, whereas the weaker medical content — drawn from enormous volumes of reddit and facebook conversations or fiction, wherever large quantities of loosely medical discussion originate — sits in a less clinically structured register? Describe your problem in those everyday terms and perhaps a different slice of training data gets pulled in, producing entirely different replies. These tools are strange to use: astonishingly capable in some moments, yet fundamentally quite stupid next to humans. Even in domains you would assume they fumble — the logical sequence behind diagnosis, or deciding which tests to order to figure something out — they can perform rather well, except there is no real capacity for that kind of reasoning; it simply emerges from the next word generator.

 

[Flash-BCR]

A few weeks back I began with ChatGPT. My job is compliance, and part of it means drafting license applications. That means dealing with a batch of documents that come in from external people, and it is incredibly handy for reading through them, organizing them...and then lining them up against earlier applications...

There was an export license I needed to file with an agency I had no prior experience with. I had GPT walk me over to the site, assist in making sense of the information and complete the online form. I would grab a screenshot of every page of the form, and it would guide me through each stage...What had been a site I never touched turned into a finished application within a couple of hours...