Memory & Personalisation Are OpenAI's Priorities For GPT-6
OpenAI CEO Sam Altman has provided rare insight into the company's plans for GPT-6, emphasising that the next iteration of ChatGPT will focus heavily on memory capabilities and user customisation.
Speaking to reporters in San Francisco last week, Sam revealed that OpenAI's forthcoming model would arrive soon and that users wouldn't have to wait quite as long as they did between the release of GPT-4 and GPT-5.
However, he did decline to reveal exactly when we can expect to see the new model.
Instead, Altman focused on what the next gen of ChatGPT would be able to do.
"People want memory," he said during briefing. "People want product features that require us to be able to understand them."
This emphasis on memory would be quite the shift in how OpenAI has gone about developing its flagship product up until now.
Rather than simply responding to questions, it looks as though GPT-6 will try to remember the unique preferences, routines and individual characteristics of each user, making it able to interact in a more meaningful and personal way.
Addressing ideological concerns
Sam indicated that future versions of ChatGPT would also comply with recent federal requirements for ideological neutrality in AI systems.
He described his vision for a model with a centrist default stance that users could then customise according to their preferences.
"I think our product should have a fairly centre-of-the-road, middle stance and then you should be able to push it pretty far," Sam explained.
"If you're like, 'I want you to be super woke' – it should be super woke."
This flexibility extends to conservative viewpoints as well, he noted, suggesting a departure from concerns about inherent biases in AI models.
People want product features that require us to be able to understand them.
Learning from GPT-5's missteps
Altman's briefing about GPT-6 comes hot on the heels of the launch of GPT-5, which came out this month. While OpenAI made several changes to the model for this latest version, its rollout has been rocky.
Lots of users have complained on social media that GPT-5 seem colder and less helpful than its predecessor.
Altman understands this, which is likely why he is turning attentions to the future already.
"I like the new one much better," he said, conceding that the launch of the latest model has been mishandled.
OpenAI has since quietly implemented a tone update to make GPT-5 "much warmer".
Privacy and encryption challenges
Despite Sam's enthusiasm for enhanced memory features – which he called his favourite addition this year – significant privacy concerns remain unaddressed.
Temporary memory in the current system lacks encryption, potentially exposing sensitive user information.
Sam acknowledged that encryption "very well could be" added to future versions, though no timeline exists for implementation.
He particularly highlighted the need for stronger protections around legal and medical queries.
"It's in society's interest for people to get good medical advice … good legal advice," he said.
Beyond chat interfaces
Looking further ahead, Sam expressed interest in brain-computer interfaces as a potential evolution of AI interaction.
He described neural interfaces as "a cool idea" and imagined scenarios where users could "think something and have ChatGPT respond".
However, he also suggested that traditional chat interfaces may have reached their limits.
"The models have already saturated the chat use case," Sam explains.
"They're not going to get much better... And maybe they're going to get worse."
OpenAI has reportedly been collaborating with psychologists to measure user wellbeing over time, though this data remains unpublished.
Sam hinted that the company might release these findings in future, as it continues to shape ChatGPT into a more adaptive and personal AI assistant.
