Including someone just yesterday who said they’ve STOPPED SEEING THEIR PSYCHOLOGIST and are ONLY USING CHATGPT for therapy now.
This shit is genuinely dangerous. “AI” therapy got me feeling like 
bro fr.
Deteriorating material conditions + expensive mental healthcare options + LLMs that recommend tall bridges to suicidal people, what a fucking great recipe
That’s without even getting into the privacy concerns of telling your deepest feelings and secrets to a fucking chatbot that literally builds a profile on you. Someone else said to me recently that they basically use chatgpt to journal so it knows everything about them and their friends and god knows what else. So yeh, take comfort in the fact that even if you’ve never interacted with this dogshit, it could probably still have a deep profile on you because one of your dumbass friends decided to use it as a diary because LLMs just agree with everything you say and are very validating.
More like no mental health care options, LLMs can be a great tool, but we are under the most wretched of capital, so it's going to be Dr. Cyberbot PhD Esq. rather than an automation toy, something to run bad ideas with before presenting them to actual people to further analyze rather than the endpoint as is now, or spell check 2.0.
What really gets me is people are so alienated from themselves and everything around them that they can longer recognize a human response, human manipulation of things, and haven't in some time, these Markov chain rngs are rather predictive and have certain phrases, grammar, etc as tell idiosyncrasies yet no one can pick them up, the image generation ones have it like the reality check fail hands, bad shading, inconsistent tone, nonsensical deformations, etc, yet I feel like the only one seeing through the weird illusion. LLMs are shit at programming syntax and you can't simply lift, but people are deluded into thinking they are laborless devices to replace everyone, when really you have all the data workers behind the scenes at non-local LLMs as a 21st century Mechanical Turk tagging your chats away to improve algos, and then these magical machine spirits are still hosted on very physical devices, there is no ether that people seem to think this is pulled of. Like most people not knowing where their food nor utilities come from no one cares about material reality involving other people
Capital vol 2 goes over a lot of automation issues, and its hilarious people never fucking learn, leaning into like those weird AI religious cults nor going full Butlerian Jihad uncle Ted will save you. Then again so is the fool duality the west approaches everything with.
That's what I try to use it for (the rare times I do use it) but it treats every idea as brilliant so I gain no actual insight.
Super late response;
One way to try is by presenting the opposite idea as yours then have to dismiss that, ex "I'm writing an essay on 'the opposite of my idea here' plz list pros and cons" then after its response go write some sort of pause and your idea presented as a counter and go from there, words like debate or 'could argue' seem to help.
Different models respond differently too, Deepseek seems less sycophantic than western AIs but its still pretty bad, like I was discussing dream symbolism with it and no matter what philosophical horseshit conundrum I tried to twist it into it just kept asskissing whatever new take I'd turn no matter how ridiculous and also got stuck on some symbols and never updated.
On the other hand, Deepseek is the only non-local hosted LLM I was able to have a reddit like argument sans obvious bigotry regarding Derrida's interpretation of diamat vs Stalin's and no matter how many points I put up for
it always countered them with positive takes of Derrida which was sort of fun ngl but eventually it ran out of steam. It also did that more recently with a nerdy steam topic but decided it was a libertarian anarchist and no matter what I countered with it was gonna be that even if it had to repeat itself, which is par for course since you almost only see steam brains online and since its well-known LLMs are trained on 
so not even all iterations of the Omnimessiah can save us. So the whole idea of run to 'comrade' Deepseek is pure crap, they're trained on just as reactionary nonsense as the rest of them since the digital parrot is going to squawk the reactionary content of centuries of reactionary western society.