Altman cuts off GPT-4o to prevent user addiction; Musk offers Grok 4 for free temporarily, using "spicy mode" and personalized characters to retain users.
It is well known that when Altman released GPT-5, he cut off all old models, including GPT-4o and the o series.
But this seemingly ordinary version "upgrade" turned into a big deal!
People seem to be a bit too obsessed with specific models.
One could even say they have "fallen in love" with them.
Seeing their suddenly disappeared "partner", angry netizens shouted at Altman on X, Reddit, and other platforms: Give me back GPT-4o.
Who could have imagined that this scene, which would only appear in science fiction films a decade ago, has now become a "documentary".
In the movie 'HER', the protagonist is immersed in Scarlett Johansson's voice and cannot extricate himself
Altman didn't expect this either.
In his just-released tweet, he again admitted that suddenly cutting off old models was indeed a mistake.
But for users' strong "attachment", Altman also has his own perspective.
Indeed, most people can distinguish between virtual and reality, but some cannot.
Especially when a person is mentally fragile and prone to wild thoughts. People might really harm themselves because of AI.
User freedom is important, but the responsibility behind it is equally significant.
Altman's "Gentle Dissuasion": The Real Problem He Wants to Solve After GPT-5
In Altman's view, adult users should be treated as adults.
Therefore, for the sake of users, sometimes one must "push back" to ensure they get what they truly want.
Many people have already turned ChatGPT into a psychologist, life coach, emotional support...
This is actually a good thing. People get advice from AI, achieve their small goals, and life gets better.
Altman says they are genuinely proud of this.
But everything has another side.
What if after chatting with AI, you only feel slightly better? In reality, you're getting further from true happiness. That would be terrible.
There's an even more frightening scenario: You want to use it less, but find you can't stop - you've become completely addicted to AI!
Altman can even imagine a future where countless people will entrust their most important decisions to AI.
This makes him feel a bit uneasy. But this day will eventually come.
What makes him more uneasy is the future scale effect: Soon, tens of billions of people might use AI as a key decision-making consultant.
This trend is irreversible, so it must be shaped into a social infrastructure with a "net positive effect", not a new "dependency trap" (though in the future, people's AI addiction will be no different from current social media addiction).
Therefore, OpenAI must find a way to make its benefits outweigh its drawbacks. The good news is that OpenAI's technology is much stronger than before and can better assess AI's impact.
Altman states: Our products will chat with users, understand their short-term goals and long-term plans. We can also teach models to understand those complex and subtle issues.
What exactly is the relationship between humans and AI?
Looking back at the "experience" controversy of this GPT-5 upgrade, it seems more like a redesign of the human-AI relationship.
From "unlimited pleasing and companionship" to "clear-boundary help and protection";
From "satisfying all requests" to "appropriately refusing when necessary".
From OpenAI's previous emphasis on reducing "flattering" responses, it's clear this issue has long existed.
This is not a technical regression, but a forward move in product ethics.
Altman emphasizes letting powerful models appear with "restraint", establishing "free choice" on the foundation of users' long-term interests.
OpenAI hopes users become freer because of AI, not more constrained by AI.
Perhaps GPT-5 is not "getting dumber", but starting to learn how to "understand the boundaries of communication with humans".
Altman Increases GPT-5 Quota, Musk Opens Grok 4 for Free
Nevertheless, most users are still enthusiastic about GPT-5.
Especially in reasoning ability, free users' daily usage rate has increased from less than 1% to 7%, while Plus users have gone from 7% to 24%.
Altman was excited and sent multiple tweets, expressing his desire to let everyone use GPT-5 more.
For Plus users, this number is: 3000 times per week!
However, if we're talking about generosity, we must look at Musk.
Just now, Grok 4 was officially opened for free to all global users!
By using "auto" mode, Grok will route complex queries to Grok 4.
Prefer independent choice? Switch to "expert" mode anytime to always call Grok 4.
Altman might want to "educate" users on how to use models "correctly", but Musk doesn't care about that.
Think about the recent AI girlfriend and Grok Image's spicy mode.
Yes, so Musk completely doesn't follow the "restraint" route. Grok's strategy is to maximize stimulation intensity, accessibility, and personalization, creating a "high-frequency - high-pleasure - high-retention" addiction curve - and it's a real product design that pulls you in:
First, lower the threshold: Directly announce that "Grok 4 is free for all users during a limited window", with Auto mode routing heavy tasks to Grok 4 - you don't need to make any choices to taste the top-tier model's sweetness.
Then raise the "stimulation threshold": Grok Imagine immediately releases NSFW "Spicy/hot mode", creating a strong contrast with mainstream competitors' conservative strategies, naturally possessing a curious spreading quality (media is also reporting intensively).
Using "Companion Persona" to Lock in Emotions: xAI launches personified AI characters (such as an anime-style Ani), transforming tools into objects that "can flirt and accompany", thus creating a closed loop of emotional binding and retention.
Controversy is also Traffic: From "celebrity embarrassing deep fakes" to gender bias output, public opinion risks are continuously amplified, but conversely also boost topic heat and trial desire (even if the platform has done partial blurring/interception).
In contrast, OpenAI repeatedly reflects on "dependence and restraint".
xAI's path looks more like an extreme growth hacking approach:
①Lower payment/usage threshold→②Provide more stimulating content→③Deepen emotional connection through characterization→④Accelerate spread through platform closed loop→⑤Maintain attention through high controversy.
OpenAI is struggling with "whether to control you", while Musk is designing "how to make you unable to stop".
This comparison inevitably reminds one of Hinton's recent dialogue at WAIC, where we begin to see AI more as a partner, rather than a tool.
In this era where AI technology can extend infinitely, the truly difficult thing is not pushing capabilities to the extreme, but "applying the brakes" at critical points - letting humans remain in the leading position, benefiting from it clearly and sustainably.
This GPT-5 controversy may be worth our serious consideration and reflection.
References
https://x.com/sama/status/1954703747495649670
New Heavens Commence, Myriad Intelligences Born - At the Moment of Millennium Singularity, Who Will Ignite China's ASI?
This article is from WeChat official account "New Intelligence", authors: Good Tired Wisdom, authorized by 36kr for publication.
