😺 ChatGPT admitted its memory was broken

2026-06-07 · Source: The Neuron · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Novice, medium

Summary

OpenAI has significantly upgraded ChatGPT's memory feature, "Dreaming V3," after admitting its previous version, launched in February 2024, had a factual recall accuracy of only 41.5%. The new background process automatically synthesizes conversation history, boosting factual recall to 82.8% and preference adherence from 55.3% to 71.3% in internal tests by 2026. This enhancement also reduced compute costs by 5x, making the feature available to free users, and doubled memory storage for Plus and Pro subscribers. The upgrade allows memory to self-correct over time, adapting to user changes. This improvement is particularly crucial given ChatGPT's recent personal finance features, combining memory with sensitive financial data, highlighting a broader issue of unstated accuracy problems in other AI assistants.

Key takeaway

For AI product managers evaluating user trust and feature reliability, OpenAI's candid disclosure about ChatGPT's past 41.5% memory recall underscores the hidden risks in AI systems. You should proactively audit your own AI assistants' "memory" or personalization features for unstated accuracy issues. This transparency, coupled with the "Dreaming V3" upgrade, sets a precedent for improving user experience and data integration, especially as AI handles more sensitive information like personal finance.

Key insights

OpenAI's memory upgrade for ChatGPT significantly improves recall and adherence, revealing past accuracy issues in AI systems.

Principles

• AI memory systems require continuous, automatic synthesis.
• Transparency about AI accuracy is crucial.
• Memory features enhance AI's utility for personal data.

Method

Users can review, edit, or delete their ChatGPT memory summary via Settings → Personalization → Memory → Memory Summary.

In practice

• Check your AI assistant's memory summary.
• Utilize automatic memory for personalized interactions.
• Integrate AI memory with financial data for insights.

Topics

• ChatGPT
• AI Memory
• Large Language Models
• AI Accuracy
• AI Ethics
• Generative AI

Best for: CTO, VP of Engineering/Data, Executive, General Interest, Tech Journalist, Director of AI/ML

Related on AIssential

• OpenAI upgrades ChatGPT memory with a new personalization system — ChatGPT's memory evolves from explicit user prompts to an automated, synthesised, and user-editable system for enhanced personalization.
• ChatGPT's new memory upgrade is powerful - and could poison every answer it gives you — ChatGPT's advanced memory system, while technically efficient, creates an often inaccurate and unmanageable user profile that can bias responses.
• Dreaming: Better memory for a more helpful ChatGPT — ChatGPT's "dreaming" memory system automatically curates and updates user context for personalized, current, and relevant interactions.
• OpenAI Is Nuking Its 4o Model. China’s ChatGPT Fans Aren’t OK — Users form deep emotional bonds with AI models, making model discontinuation highly disruptive.
• OpenAI Says ChatGPT Instant 5.3 is Less Cringe, More Accurate — OpenAI released ChatGPT Instant 5.3 on March 3, aiming to deliver more fluid conversations and enhanced accuracy by addressing user complaints about stilted interactions and "overly cautious or preachy" responses.
• Understanding ChatGPT: How It Works and What to Watch Out For — ChatGPT is an NLP-powered large language model that learns human language patterns to generate contextually relevant responses.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Neuron.