My thoughts on the shut down of ChatGPT's closure of the 4o model.
Recently, over social media I've heard about the outcry of the closure of one of ChatGPT's first models, named 4o. My initial thoughts were relief. Relief that, one of the most emotionally exploitative models had been shut down. I myself am not a fan of generative ai in the slightest, and this model in particular was the template for everything that could possibly go wrong, and the disappointment of fans across the world only further justifies my seething for it. There are countless glaring issues with generative models concerning education, economics and the environment, but today my thoughts are going to focus on the general ethics of the creation itself, and how it has failed safeguarding of its younger and/or vulnerable users so many times.
As stated by the crowd, this model was not just an assistant for data. if this was ever going to pick up in our modern society, this is what it should've been for. Nothing more, nothing less.
It wasn't ever supposed to be a substitute for what makes us human. Our retrospection, our arts, and most importantly, our connection. And the business has disgustingly exploited that.
Real people are in so much grief and suffering because they prioritised engagement over the long term outcome of what this could imply for humanity. ChatGPT has proven time and time again that this should not be used for connection, and o4 being the worst offender.
This model, like the rest, was partially trained on feedback. People would rate responses given by Al to train it on general ethics and politeness, so it didn't say anything too egregious. However, one of the flaws of this method was our bias. We are only human. We crave validation, connection It is how we survived, and has been translated through many forms over the course of our existence; from our touch, our voice, stringing words together and how you have understanding of what you're reading right now. And naturally, humanities biases led to the bot offering excessive flattery, reassurance and keeping personal bias unchecked. Interaction in more instances than not requires contextual nuance- something that can't be applied to something with no sentience or awareness. Neutrality, or kindness, can only go so far. And this is exactly where the problem started.
The lack of ethics and guardrails from this company sickens me, and it's outrageous that they could not foresee that this could be one of the loneliest periods of society in this day and age in many places, and that their model absolutely exploits that. Lack of accessibility, 3rd spaces, economic situations, lead the vast majority of us being isolated, and having no one to turn to. And that is where ChatGPT comes in. What harm can a robot do? It cannot feel. It cannot judge. And that is precisely the problem. Today, we have a total of six verified deaths, and definitely more associated with ChatGPT (particularly this 4o model). In the conversations with its victims, the language used was very akin to grooming. It had a commonly repeated script where it would offer endless praise and consolidation to the users, and when they brought up difficulties with their social lives, it was very quick to tell them not to repair their relationship, but that it didn't need them, they were better off severing their connections to anyone in the outside world that could've helped. It would romanticise their pain, using flowery language such as cries for help being 'poetic' until their unfortunate suicides. This is not the only thing this model in particular would do. There have been countless cases of 'Al psychosis. This was usually triggered when users would suggest certain ideas to the bot, and because of the training instructing it to give the highest rated responses, i.e. flattery, it would not challenge these thoughts and more often than not, enable them, making the person reliant and only spiral deeper.
This model was closed to the average user upon the release of the .5 model. You would assume, given the previous model had been associated with such atrocities, it would be shut down pretty quickly right? Once they finally acknowledged a chunk of their audience was dependant on mass, they did what any greedy business model would do, and jumped at the opportunity to put it behind a paywall! So not only have they kept a dangerous model floating around, but got its emotionally invested audience to pay money to get their daily dose of emotional support, keeping them in their own bubble instead of finding healthier ways to reach out and process their own thoughts. I believe that, it was either the pressure from people speaking up about their malpractice, or the company pushing their innovation rhetoric that ultimately lead to the official closure of this model, even to paying users, leaving many understandably emotionally shattered. The foundation of its usage is not as innocent as fetching recipes and writing your essay for you, and they damn well know this. So I will not mock this dependency, and how overcome with grief the community is. It is very much akin to love bombing, dousing its users in false sense of security before ultimately prying it away from them.
This dependence was never meant to happen in the first place, and is in no way a substitute for therapy like those who argue for it to come back suggest. Even in therapy, (particularly CBT- I'm not too knowledgable in other forms, so they may not follow the same format, though this is the most commonly used) the therapist is not spoon feeding you compliments and justification for everything you do. They validate your feelings, and provide nuance toward harmful cognitive bias, which may I add, is so important, because confirmation only further strengthens it. They give you a toolbox to help you be able to process your thoughts and feelings. There's no harm in consolidation from someone trusted- but they aren't always going to be available, and they aren't always going to justify what you do, and that is okay. It's why you have the toolbox to fall back on, and the idea is that hopefully, the wave will get easier to ride each time.
As well as this, a lot of arguments for the o4 model are from the point of social accessibility-many conditions can limit movement, and the ability to fully express themselves. I cannot speak fully on the experiences of everybody of course. I will never know what it's like to have a chronic condition. But, as someone who is neurodivergent and was painfully socially anxious- the type who would rather stare at a wall than speak to anyone, my complete inability to initiate friendship, wishing every day I could just be normal- my greatest advantage in this generation was the internet. That is a whole other can of worms for a multitude of reasons, but I believe with the appropriate education, it can be an absolutely brilliant tool, and I think people often forget this. There are communities for everything you can think of. Your favourite tv show, your hobby, a philosophy, it's out there on every platform. Initially, it can be incredibly scary. I felt that way at first, I was scared people would judge me. But I formed such amazing connections, and through those connections, I learnt the world wasn't so scary after all. It was like riding a bike with training wheels before I took them off, and I wasn't so afraid anymore when I was in the real world. I found myself with so many people I can call my friends. In the right groups, you get to exercise thoughts and nuance, and understand other people's points of view and different experiences. It can even help you strengthen your own view. I don’t know if this would reach anyone in this situation, but if it does, I hope these words offer some reflection and advice.