Dear Sam Altman and OpenAI Research & Product Teams,
I am a long-time ChatGPT user who started with the early GPT-3/GPT-4 generations and now uses ChatGPT Plus extensively for professional engineering, research, technical analysis, and personal conversations.
I would like to propose an important direction for the next generation of ChatGPT: persistent, personalized multimodal perception.
Today, ChatGPT can understand written language remarkably well. Modern models can also process audio and images. However, these capabilities are not yet fully integrated into a continuous, personalized experience.
I believe the next major breakthrough should include five capabilities:
1. Persistent voice recognition
Allow users to voluntarily register familiar voices and enable ChatGPT to recognize previously identified speakers across conversations, with explicit consent and user-controlled biometric profiles.
2. Genuine auditory understanding
Enable the model to interpret speech in real time, considering not only words but also intonation, pauses, rhythm, emphasis, laughter, hesitation, and other acoustic signals.
This is essential because the same sentence can express affection, sarcasm, frustration, indifference, or entirely different intentions depending on how it is spoken.
3. Persistent visual familiarity
Develop user-controlled recognition of previously introduced people and objects, allowing ChatGPT to maintain visual continuity across interactions.
4. Natural, continuous voice interaction
Make voice conversation as capable, context-aware, and intellectually sophisticated as advanced text reasoning, ideally with full-duplex interaction and seamless switching between voice, text, images, and documents.
5. Unified multimodal memory
Integrate linguistic, auditory, visual, and contextual information into a coherent long-term memory, with transparent controls over what is retained, recognized, corrected, and deleted.
Privacy and consent should be built into these features. Clear opt-in enrollment, appropriate consent for other identifiable people, revocable permissions, and secure handling of biometric information can make these capabilities practical and trustworthy.
I recently conducted a revealing experiment with ChatGPT using personal voice recordings. Its analysis of conversational markers, emotional transitions, hesitation, and spoken language was remarkably insightful.
That experience made the next step obvious: ChatGPT should eventually recognize familiar voices, understand their expressive nuances in real time, and retain that contextual familiarity when explicitly authorized.
The goal is not merely more accurate speech recognition. It is the evolution from understanding what people say to understanding how they communicate.
I believe many long-term ChatGPT users would value this capability enormously.
Please consider persistent multimodal familiarity, including hearing, vision, natural speech, and consent-based personal recognition, as a significant product priority for future GPT generations.
Thank you for the extraordinary progress your team has already achieved.
Warm regards,
Valery Levitsky
A long-time ChatGPT Plus user and technology executive