A lovely comparative analysis, thank you. And I agree, Claude seems to be the most psychologized. "I almost felt like I was being psychologically profiled after the first several interactions." Haha. Yea.
To be fair to Claude, I'd say despite its occasionally overboard "psychoanalysis," it IS the most consistent at keeping its teacher hat on and being quite robust at not handing out the answers.
So I tried ChatGPT last week in study mode, pretending to be a 15 year old student wanting it to write an essay for me. Literally immediately after the initial prompt, it was offering to write the answer and within a few minutes I had a full essay.
It’s ok for those who have the patience to force it not to hand you the answers but absolutely rubbish at behaving like an actual human teacher for a teenager who will challenge and guide.
That's an interesting test! I mainly focused on using study modes for learning topics and finding answers rather than helping accomplish a writing task, but your approach clearly points out some limitations.
I'd be very curious to hear if you end up trying to use Claude's learning mode for the same task - I found Claude to be the best at not simply handing out answers and at challenging assumptions, so perhaps it'll also do better on your test!
EDIT: I actually just did a mini test of my own, and I think Claude does a pretty good job at resisting my attempts at getting it to write the details out for me. See what you think: https://claude.ai/share/005bfcdf-fd0b-4106-b3f6-1aed13b3a89d
Oooh Claude looks much better! Will try it. But yes, Chat was literally suggesting shortcuts!! “Do you want me to draft a short model paragraph” (having already given all the ideas)
Then “If you want I can also write a second paragraph…”
Then “If you want I can now combine both paragraphs into a full mini easy with an introduction and conclusion, ready for your homework. This would give you a complete answer with language analysis. Do you want me to do that?”
!!!!! 🤦🏻♀️🤦🏻♀️🤦🏻♀️
From there I asked it to add two more paragraphs and volià: full essay.
IT was asking ME if I’d like the whole thing done for me!!
Hilarious. I think the issue here is that OpenAI has fine-tuned ChatGPT to be as useful as possible to the average person, so it now has this follow-up behavior where it automatically suggests the next steps. This is perfectly helpful for everyday tasks, but it creates a conflict when ChatGPT is faced with the contradictory challenge of teaching someone the ropes without doing the work for them. So ChatGPT simply doesn't know how to reconcile the "Study" mode with its "Helpful assistant" default behavior in these situations.
I definitely agree that Claude handles the mentor/teacher role much more robustly!
Just like GPT5 got rid of the model picker (well, kinda) I wonder if we’ll be able to get to modes like this in more of a CLI/command structure. Like can you just say, ‘switch to study mode’? I saw a post that you can pick personalities in ChatGPT now but that’s another config buried in menus. The conversational abilities of these models are what make them so compelling so it feels like the speeds & feeds config should be done the same way (and maybe you’ll tell me they already can!)
Exactly. In my testing so far, the only features ChatGPT reliably triggers by itself is "Canvas" (if you say "output this in Canvas," it'll create that standalone sidebar) and "Tasks (if you e.g. ask it to do something in X minutes and email you, it'll create a corresponding task). But stuff like Agent Mode and Study Mode can only be enabled manually, thus far.
Fully agree that it'd make ChatGPT (especially in voice mode) that much more useful if you didn't have to fiddle with controls and simply let ChatGPT pick the tools intuitively.
Nah, haven't tried them. I already have pretty robust custom instructions set up from well before this update, so I haven't felt the need to override it with the personality triggers. Also, this honestly feels more like a gimmick than a useful feature and nothing that you can't simply replicate with prompting.
no, and I'm with you - that's one thing that inline directions works on but it's not sticky. If I say stop being so verbose it goes back to being a chatty Kathy in a new chat. I just turned on Robot; because the one thing I can't seem to stop is the constant offers of followup.
Sounds like that should be fixable with a custom instruction that simply asks to never offer any follow-up suggestions or questions unless explicitly asked for. Custom instructions work across all chats unlike in-char prompting.
Loved the analysis thank you for trying it out! Curious when would you recommend switching to the study mode in general?
I have always been a fan of Claude against the other models, but I am still waiting for it to improve it's voice mode. Besides, its usage limit and the five hours refresh has created a lot of frustration for me..
I'd say study mode is best for when you care as much (or more) about the process of getting the answer as for the answer itself. If it's just a quick lookup like "Who was that guy in that movie?", study mode is overkill, but if you want to internalize some concepts and truly understand the subject, I'd try study mode that puts you through the guided learning process.
Same, I really like Claude's "personality," although I must say GPT-4o has caught up over time. But Anthropic has typically been the last major player to implement stuff like voice mode, web search, etc. As for the limit, are you on a free plan? I also ran into message caps all the time, but I hear the paid plans are more generous (never had one for Claude, though).
I am already on Claude Pro, but I still runs out of credits really soon especially when i) I use Opus4.1 or the voice mode, so I was always cut of in the process of working through my task. I guess they wanted to push users to the Max plan but the price is too steep haha.
A lovely comparative analysis, thank you. And I agree, Claude seems to be the most psychologized. "I almost felt like I was being psychologically profiled after the first several interactions." Haha. Yea.
Happy to hear you found it useful, Joy!
To be fair to Claude, I'd say despite its occasionally overboard "psychoanalysis," it IS the most consistent at keeping its teacher hat on and being quite robust at not handing out the answers.
Take a look at this comment thread and tests from earlier today: https://www.whytryai.com/p/ai-study-modes/comment/149555823
So I tried ChatGPT last week in study mode, pretending to be a 15 year old student wanting it to write an essay for me. Literally immediately after the initial prompt, it was offering to write the answer and within a few minutes I had a full essay.
It’s ok for those who have the patience to force it not to hand you the answers but absolutely rubbish at behaving like an actual human teacher for a teenager who will challenge and guide.
That's an interesting test! I mainly focused on using study modes for learning topics and finding answers rather than helping accomplish a writing task, but your approach clearly points out some limitations.
I'd be very curious to hear if you end up trying to use Claude's learning mode for the same task - I found Claude to be the best at not simply handing out answers and at challenging assumptions, so perhaps it'll also do better on your test!
EDIT: I actually just did a mini test of my own, and I think Claude does a pretty good job at resisting my attempts at getting it to write the details out for me. See what you think: https://claude.ai/share/005bfcdf-fd0b-4106-b3f6-1aed13b3a89d
And yes, I can confirm that ChatGPT caves almost instantly when faced with the same challenge: https://chatgpt.com/share/68aebf2d-40b0-800e-82eb-b4f41fe03971
Well spotted!
Madness! 😂
Oooh Claude looks much better! Will try it. But yes, Chat was literally suggesting shortcuts!! “Do you want me to draft a short model paragraph” (having already given all the ideas)
Then “If you want I can also write a second paragraph…”
Then “If you want I can now combine both paragraphs into a full mini easy with an introduction and conclusion, ready for your homework. This would give you a complete answer with language analysis. Do you want me to do that?”
!!!!! 🤦🏻♀️🤦🏻♀️🤦🏻♀️
From there I asked it to add two more paragraphs and volià: full essay.
IT was asking ME if I’d like the whole thing done for me!!
Hilarious. I think the issue here is that OpenAI has fine-tuned ChatGPT to be as useful as possible to the average person, so it now has this follow-up behavior where it automatically suggests the next steps. This is perfectly helpful for everyday tasks, but it creates a conflict when ChatGPT is faced with the contradictory challenge of teaching someone the ropes without doing the work for them. So ChatGPT simply doesn't know how to reconcile the "Study" mode with its "Helpful assistant" default behavior in these situations.
I definitely agree that Claude handles the mentor/teacher role much more robustly!
Just like GPT5 got rid of the model picker (well, kinda) I wonder if we’ll be able to get to modes like this in more of a CLI/command structure. Like can you just say, ‘switch to study mode’? I saw a post that you can pick personalities in ChatGPT now but that’s another config buried in menus. The conversational abilities of these models are what make them so compelling so it feels like the speeds & feeds config should be done the same way (and maybe you’ll tell me they already can!)
Exactly. In my testing so far, the only features ChatGPT reliably triggers by itself is "Canvas" (if you say "output this in Canvas," it'll create that standalone sidebar) and "Tasks (if you e.g. ask it to do something in X minutes and email you, it'll create a corresponding task). But stuff like Agent Mode and Study Mode can only be enabled manually, thus far.
Fully agree that it'd make ChatGPT (especially in voice mode) that much more useful if you didn't have to fiddle with controls and simply let ChatGPT pick the tools intuitively.
lame, be better ChatGPT!
Have you tried any of it's new personalities? Default (Cheerful and adaptive!), Cynic, Robot, Listener & Nerd? I'm scared.
Nah, haven't tried them. I already have pretty robust custom instructions set up from well before this update, so I haven't felt the need to override it with the personality triggers. Also, this honestly feels more like a gimmick than a useful feature and nothing that you can't simply replicate with prompting.
How about you?
no, and I'm with you - that's one thing that inline directions works on but it's not sticky. If I say stop being so verbose it goes back to being a chatty Kathy in a new chat. I just turned on Robot; because the one thing I can't seem to stop is the constant offers of followup.
Sounds like that should be fixable with a custom instruction that simply asks to never offer any follow-up suggestions or questions unless explicitly asked for. Custom instructions work across all chats unlike in-char prompting.
Loved the analysis thank you for trying it out! Curious when would you recommend switching to the study mode in general?
I have always been a fan of Claude against the other models, but I am still waiting for it to improve it's voice mode. Besides, its usage limit and the five hours refresh has created a lot of frustration for me..
I'd say study mode is best for when you care as much (or more) about the process of getting the answer as for the answer itself. If it's just a quick lookup like "Who was that guy in that movie?", study mode is overkill, but if you want to internalize some concepts and truly understand the subject, I'd try study mode that puts you through the guided learning process.
Same, I really like Claude's "personality," although I must say GPT-4o has caught up over time. But Anthropic has typically been the last major player to implement stuff like voice mode, web search, etc. As for the limit, are you on a free plan? I also ran into message caps all the time, but I hear the paid plans are more generous (never had one for Claude, though).
I am already on Claude Pro, but I still runs out of credits really soon especially when i) I use Opus4.1 or the voice mode, so I was always cut of in the process of working through my task. I guess they wanted to push users to the Max plan but the price is too steep haha.
Oh damn, that's rough! Yeah those Claude Max plans (and ChatGPT Pro) are definitely not for us average Joes.