Socratic questioning
Turn an AI from something that explains things to you into something that makes you explain things to it, and find out, in about ten minutes, which parts of a topic you only think you understand.
Understanding shows up in questions, definitions, and defences, not in nodding along.
Here you argue, steelman the view you dislike, force a precise definition, and hunt for the case that breaks your rule.
Especially useful in philosophy, literature, history essays, and anywhere a marker writes “assertive but thin.”
Turn an AI from something that explains things to you into something that makes you explain things to it, and find out, in about ten minutes, which parts of a topic you only think you understand.
Be led to an idea by evidence and questions until you produce it yourself, rather than being told it and then shown why it's true.
Attempt a problem beyond your current ability before instruction, so the lesson lands on gaps you already felt, instead of on a blank slate that feels clear and teaches less.
Defend a claim under sustained, specific questioning from someone who is trying to break it, and find out which parts of your position are load-bearing and which were ornamental.
Have cases produced that your understanding doesn't cover, and find out that the rule you'd have stated was the wrong shape.
Find out where a correct rule stops being useful, and what takes over.
Before settling on a conclusion, spend two minutes arguing against it.
Put the source in the witness box before you cite it.
Decide which of two conflicting sources is stronger (and why) before seeing commentary that tells you what to think.
Defend a position against deliberate attack, and find out which parts of it you hold for reasons and which you hold by habit.
Build the strongest version of a position you think is wrong (well enough that someone who holds it would recognise it) and find out what your own view actually rests on.
Have the unstated premises in your own reasoning named, the things you're relying on without having noticed you decided them.
Be made to say exactly what you mean by each vague word, until there are none left.
Restate someone's argument as numbered premises and a conclusion, so you can see what would actually have to be true, instead of arguing with a vibe.
Hold a correct position against a confident challenge, without folding to tone, and without defending errors you should concede.
Answer "why" repeatedly until you hit bedrock or failure: exposing how deep your understanding actually goes.
State a premise, then generate three independent consequences and defend each: training forward inference instead of only recognising conclusions you've already seen.
Turn a vague claim into a measurable or observable statement before continuing, so arguments stop hiding in unfalsifiable fog.
Revise work while responding in real time to continuous critique: training edit-under-fire instead of only polishing in calm after a single feedback dump.
Receive simultaneous critique from several distinct critical personas and reconcile them: training judgment under conflicting advice, not obedience to the loudest voice.
Find the flaw in arguments built to contain one, naming the failure mode precisely enough that you don't just shout "fallacy" at everything you dislike.
Have your own questions assessed and sharpened, so you stop asking prompts that only produce summaries and start asking ones that force thought.
Reason jointly with an AI that is required to disagree when warranted, so "collaboration" doesn't collapse into the model doing the thinking while you nod.
Argue each side in turn and then judge, so your final view has survived contact with the case you didn't want to make.
Work thesis → antithesis → synthesis on a contested question, ending with a position that kept what was true in both, instead of a mushy middle or a stubborn first take.