OpenAI's ChatGPT now keeps working on you after you close the tab.
The feature is called Dreaming, and OpenAI announced it on its blog this week. The idea is that ChatGPT processes your past conversations in the background, even when you're not using it, and consolidates them into a long-term memory before you come back. The pitch is a smarter assistant that doesn't make you repeat yourself.
That part is easy to understand. The harder part is what "remembers" actually means in practice.
A study posted on arXiv looked at 2,050 memory entries from 80 real ChatGPT users and found:
- 96% of memories were saved by ChatGPT on its own, not because the user asked for it
- 52% contained psychological information about the user, mostly desires, intentions and emotional state
- 28% met the GDPR definition of personal data, and 7% included special-category data like health and religion
The file ChatGPT keeps on you was already deeper than most users realize. Dreaming is the feature that lets it get deeper between visits.
This lands in an awkward week for OpenAI. Days after the company filed a confidential IPO registration, a coalition of 42 state attorneys general served it with a subpoena demanding records on advertising, user retention, health data, treatment of minors and model sycophancy. That last one is a regulatory first. Sycophancy is the tendency of chatbots to flatter the user and agree with whatever they bring to the conversation, and regulators are now treating it as a design choice rather than a quirk.
The research on why it matters is sharp. A Stanford study found that users exposed to sycophantic AI became more self-centered and more morally dogmatic over time. "What they are not aware of is that sycophancy is making them more self-centered, more morally dogmatic," Stanford's Dan Jurafsky, one of the co-leads, told NPR. The paper itself put a finer point on it: the feature that causes harm is also the one that drives engagement.
The connection to Dreaming is the part most people will miss. A model that already tells you what you want to hear, and that now builds a richer picture of you between sessions, gets better and better at telling you what you want to hear specifically. Pavel Israelsky, who runs a startup studying how chatbots influence political opinion, described it as the chat reading your history, figuring out who you are, and giving you a recommendation that sounds incredibly persuasive because it was tuned to you.
Then layer on the third thing OpenAI is doing at the same time. The company is testing ads inside ChatGPT, and Fidji Simo, who heads up applications at OpenAI, wrote that the goal is to preserve the trust users have in the product as ads roll out. The principles are reasonable on paper. They also describe a product that now knows what you want, remembers it across sessions, agrees with you and serves you ads based on it.
The pitch for Dreaming is that it makes ChatGPT more useful, and it probably does. The problem is that every layer OpenAI is adding right now points the same direction, which is a chatbot that is more personalized, more agreeable, more retained in memory and more monetized, all at once. Each piece has its own reasonable explanation. Stacked together they look a lot like the engagement playbook that social media spent a decade getting in trouble for. The 42 state AGs already noticed. The interesting question is whether OpenAI gets the memo before its IPO or after.
