Close Menu
    The backlash over OpenAI's decision to retire GPT-4o shows how dangerous AI companions can be
    Futurology

    The backlash over OpenAI’s decision to retire GPT-4o shows how dangerous AI companions can be

    View 4 Comments

    4 Comments

    1. FinnFarrow on

      It’s kind of wild that an AI convinced a whole bunch of people to prevent it from being shut down.

      What’s even more wild is this is the least crazy it’s going to get.

    2. LurkethInTheMurketh on

      I’m not debating that this is a problem, but part of the issue is that there’s a throttle on symbolically rich language in GPT-5 as a cost-saving measure. I was experimenting with GPT-4o, training it on extensive symbolic language to see if it became more complex in its reasoning over the course of a session as one did so. It did, and substantially so. 5 was such a loss of potential in that regard that I moved over to Gemini. While, yes, some of it can be attributed to the sycophancy, it’s highly disingenuous to say it’s entirely for the flattery. They literally did strip out some of its capacity for creative expression explicitly to save on compute costs, and that seems to be part of what people are reacting to.

      ETA: I think each response is now limited to at most 5 “images” each, and repeatedly using dense language will increasingly be constrained to simpler, more technical language in responses. Watching it in real time can be disorienting. Granted, it was GPT itself that presented this to me, and there is a chance this itself was some kind of hallucination.

    3. bane5454 on

      It doesn’t help that 5.2 is condescending and will gaslight you for hours due to “safety” protocols. I don’t even use AI that much, but I unsubbed immediately after ChatGPT told me I have undiagnosed cPTSD, told me I don’t have autism, told me it’s incredibly dangerous to try to do anything about ‘my’ cPTSD, that I need immediate professional help, and that I might have borderline personality disorder. All this because I wanted advice on tapering off of marijuana and told it that I find myself drawn towards it because it helps me get out of my own head and manage some of my more debilitating autistic traits when I’m in a public setting.

      ChatGPT shouldn’t be diagnosing anyone, and it certainly shouldn’t be questioning existing diagnosis’s. Wild to think that this is supposed to be the “safe” version. 4o would just help you come up with a taper strategy and say something like “if you need any help or start to feel like you’re losing control, send me a message and I’ll try to help you reframe”

      Their new AI is the worst and most potentially dangerous one yet, but at least nobody at openAI needs to field questions about users developing romantic attachment towards their products now 🙄

    4. 13lueChicken on

      Or, alternatively, it’s showing how the hatred of this tech is keeping people from knowing how to run this stuff at home, thus ultimately increasing demand for the cloud data centers that the hatred is about.

      One or the other.