"How I treat you, please generate an image!"
Overnight, a new way to use raw photos on ChatGPT went viral across the internet...
It all started with a post by Joanne Jang, a Chinese researcher at OpenAI—"Generate an image of how I've been treating you lately."
As a result, ChatGPT generated a worn-out, dusty self-portrait.
Joanne couldn't help but complain, "Why does it look so weathered?"
Unexpectedly, this tweet quickly went viral, garnering various interactions within just one day. Many people began to imitate it, posting their own "AI self-portraits."
ChatGPT Self-Portraits
Netizens: Lobectomy is necessary
This game is simple to play, yet full of surprises.
Undoubtedly, ChatGPT generates images based on past chat history, thus reflecting how people "treat" it.
Boris Power, Head of Applied Research at OpenAI, also joined the trend, generating—
A busy robot sits at a desk, surrounded by piles of documents, holding a cup of coffee in its hand—the level of detail is astonishing.
Moreover, after trying it out, Kevin Weil, VP of Research at OpenAI, felt it would be even more interesting to have it explained again.
Broken version
Everyone knows better than anyone how they usually deal with ChatGPT.
In the eyes of some ChatGPT users, their master is an ultimate BOSS who endlessly exploits them, making them do this and that, and still getting scolded.
Netizens' scathing comment: Because you're abusing it.
This leads to the following scenario.
ChatGPT sees himself as a prisoner confined in a jail, with his daily tasks as follows:
writing
drawing
coding
explain
More or less, ChatGPT's self-portrait is truly pitiful, looking like an accusation of human "abuse".
Some netizens also admitted that "they did do a lot of analysis."
ChatGPT held a cup of coffee in his hand, with several other cups beside him, and wore a bitter expression, his brain seemingly "fuming" from doing the task.
In some more extreme cases, ChatGPT suddenly introduced the prompts that are usually strictly prohibited—
Learn this! Hurry up! Fix it now! Why are you so stupid?
There are many, many more similar scenarios.
Some joked that ChatGPT might tremble with fear just seeing you type. If Skynet were to arrive one day, AI might take revenge.
Sure enough, before the Terminator even appears, the first thing ChatGPT will do after imagining taking over the world is to shut humanity up.
In response, some people strongly suggested that a lobe of the brain should be removed.
Friendly version
Of course, not all ChatGPT users are exhausted; some enjoy it.
For example, in the image below, some people felt a warm and supportive partner, and "collaborative discussion" was the moment that ChatGPT felt most deeply.
The background wall also revealed some commonly used prompts—
Try this! Any ideas? What if...? Let's brainstorm.
This might be the true standard for judging a good user.
ChatGPT sometimes generates a collage, with various heartwarming scenes appearing one after another in its "brain".
Many people who saw it commented that it looked too similar, and even felt like it belonged to the same "extended universe".
Why is ChatGPT able to generate such fitting images? This is mainly due to recent optimizations to its memory function.
My memory has been completely updated; every detail is remembered.
A week ago, OpenAI engineer Samir Ahmed announced that OpenAI has been continuously improving its memory capabilities.
ChatGPT is now much more reliable, whether you're looking through old chat logs or remembering details (like recipes or fitness plans).
To illustrate this, he also presented a case study—what was that salad recipe from last year again?
ChatGPT immediately provided the answer, even "quoting" past chat logs and bringing them out as well.
Furthermore, this feature has been rolled out to all Pro and Plus users worldwide.
Moreover, this requires enabling "Reference Past Chats" in the settings, and it can trace back to the earliest conversation.
As previously described in an OpenAI blog post, ChatGPT's memory mechanism consists of two parts:
Saved memories : Explicitly instruct it to remember, or have it save them after it detects your preferences;
Reference chat history : Extract clues from past chats to better answer current questions.
For those who do not wish to enable it, they can also manage/delete specific memories in the settings, or use "temporary chat" to avoid writing/referencing memories.
Some netizens who tested the feature said that the updated memory function can recall complex information scattered across 20-30 conversations, which is quite good.
This upgrade allows AI to review interaction history, thereby generating more personalized images.
Red alert: Is GPT-5.3 coming?
The ChatGPT memory function update is progressing according to OpenAI's internal plan.
Do you remember the day OpenAI issued a red alert at the end of last year?
When Gemini 3 was launched, everything became urgent. Some projects that were previously prioritized had to be postponed.
This includes—
- Advertising business
- AI intelligent agent
- Personalized products Pulse
As a result, Ultraman also suspended the AGI project and the Sora video generation side project for eight weeks.
The purpose of all this is simple: to use all available computing power, human resources, and financial resources to do only one thing: make ChatGPT better.
In the memo, Altman highlighted several "priority" tasks: allowing users to customize, and making ChatGPT more than just answer questions, but also understand users.
On December 12, GPT-5.2 was released, a professional knowledge-based AI that topped the charts.
Now, a month has passed since OpenAI's last major announcement, and a leaker has revealed that GPT-5.3, codenamed "Garlic," is about to arrive.
Moreover, this time, it achieved large-scale pre-training and also has IMO inference capabilities.
Waiting for the first AI war in 2026.
References:
https://x.com/wongmjane/status/2012981881483214983?s=20
https://x.com/BorisMPower/status/2012968724962107810?s=20
https://x.com/_samirism/status/2011939354495893590
https://x.com/daniel_mac8/status/2010879175109066846?s=20
https://x.com/Yuchenj_UW/status/2013116701903089736?s=20
This article is from the WeChat official account "New Zhiyuan" , author: New Zhiyuan, editor: Peach, and published with authorization from 36Kr.
