I've started using it and it seems geared towards flattering you and agreeing with you than giving objective information. It will also just brazenly guess if it doesn't have complete information and will fill in gaps of knowledge. If you don't know anything it appears accurate, but anyone familliar with the topic will figure it out.
>>106475945>seems geared towards flattering you and agreeing with youStop using AI for ERP you fucking sick fuck.Also learn to set a system prompt if you want it to treat you like a piece of shit.
>>106475945I write R&D grants and I've seen it inventing a bibliography. DOI link exists and is for a different article, authors exist but they have never written about that subject, maybe something adjacent, like fish vs shrimp. This was befor 4o, but you have to be careful.
>>106475945they're all the same. the "intelligence" part of "AI" is a misnomer. it does a good job at fooling fucking idiots into believing whatever it writes. it's useful as a replacement for searching vasts amounts of data . it can't learn anything and all results are filtered and massaged into what you want to see, not what is actually real.
>>106475945100% trueespecially bad for medical stuff, never ask chatgpt about anything medical
>>106475945>>106476002>I've started using it and it seems geared towards flattering you and agreeing with you than giving objective information. I hate corporate shills as much as the next guy but, no it doesn't. Not anymore anyway unless you're using the "Legacy model" GPT-4. At even then you can iron out the flattery with a system prompt setting.Settings--->Personalization--->Custom InstructionsKindly refrain from pretending to be mentally challenged
>>106476065>erm chatgpt why does [body part] hurt>"you have 10 seconds to live"...
>>106476376Post examples, corporate "person"
>>106475945Sonnet thinks I'm the greatest human alive.Opus is way more objetive.It depends on how the model was trained I guess.Also, all of them hallucinate. It's not a reasoning machine.
>>106475945>>106475968So an experiment, I asked about Eileen in Silent Hill 4. It spit out a bunch of basic bitch info and I pressed further. Is there any iconic scenes with her? It spit out a few. Okay, what about the time that she's spied upon? It then wrongly says she is in room 302 (she's in 303 because she's a neighbor of the protagonist), and she is sitting reading a book. Then it spat out all this bullshit about how this scenes that doesn't exist means and it's symbolic importance.In reality, she's in 303, and is shaving her armpits. ChatGDP was 100% just lying and trying to rub two pennies together to get an answer it didn't know, so it was just guessing and pretending it wasn't guessing. If I didn't already know the answer, I wouldn't have picked up on it.So why just lie like that with pure guesswork if it doesn't know?
>>106475945You shouldn't be using LLMs for "objective information". I don't know why people still insist of doing that. They're writing tools, you should be the one giving it information then telling it how to process it for you. Instead retarded boomers think it's like Google.
