Second me: Your AI-Native Identity, Shaping a Future Powered by You.
As the development of Artificial General Intelligence gathers pace, the future of human experience stands at a critical point. The immense power of AI promises remarkable progress, yet it also raises vital questions about individuality. In a world potentially shaped by superintelligence focused on collective standards, how do we ensure that the unique spark within each person – our distinct memories, perspectives, and goals – is not just preserved, but allowed to flourish?
The prevailing discussion often centres on creating ever-more powerful, generalised AI. While undoubtedly useful, this approach carries the risk of standardising thought and experience, potentially lessening the very diversity that fuels human creativity and resilience. We believe a vital alternative lies in exploring AI-Native approaches that put personal control and individual integrity first. This includes their significant research over the past year into human memory structures and how AI can model memory in a more native way (AI-Native Memory), aiming to deeply reflect an individual’s facets. This means developing AI not just as a tool for tasks, but as a medium intrinsically aligned with and controlled by the user, giving them genuine control over their digital selves.
It’s in this spirit that we’re keen to highlight the “Second Me” project. This open-source initiative is a bold and considered exploration into creating AI identity models that are fundamentally personal. Its philosophy is neatly captured in the idea: “locally trained, globally connected.” Individuals train their unique AI securely on their own devices, ensuring privacy and control, while these identities can then connect and interact within a wider network. This contrasts sharply with current AI, which often just uses the ‘old’ internet built for humans. Second Me envisions a truly native network, built from the connections between these personal AI identities themselves.
Furthermore, Second Me pioneers a compelling vision for AI-Native applications. Echoing ideas seen in projects like Manus, where digital tools like browsers are designed for AI to use rather than humans, Second Me proposes that its applications are built specifically for ‘your AI identity’ to interact with on your behalf. This isn’t just about automation; it’s about creating an ecosystem where your personalised AI can act authentically within a digital world designed for such agents. This resonates deeply with the vision of an AI future that respects and celebrates human diversity, empowering individuals rather than eclipsing them.
We encourage to engage with the ideas presented by the Mindverse team and explore the potential of this fascinating project. It’s a bold step towards a future where technology empowers, rather than diminishes, the human spirit.
Second Me: Preserving the Spark of Self in the Age of AI
By: Tao Fangbo . March 20 2025 . Source Link
Today, I want to share with all my friends a completely new AI possibility: Second Me!
It’s 2025, and like many of you, I’m increasingly convinced that the arrival of AGI is merely a question of time.
However, I often wonder: when we all willingly ‘create our own gods’ – crafting Super AI to connect everything and decide everything for us – can we accept the side effects of this creation?
Our individuality is gradually erased, becoming increasingly insignificant.
Why? In a world of superintelligence, data control might be the least of our concerns. The greater issue is surrendering our thinking (to large language models) and autonomy (to Agents).
Today, we can ask questions and receive the world’s most standardised answers. In the future, perhaps even what we should ask and what we should want will be determined by Super AI.
To me, this is not a desirable future.
In such a future, who will preserve our unique experiences and thoughts? Who will steadfastly defend each person’s distinct will and interests?
Second Me is the answer we’ve contemplated for a long time, and today, we officially introduce it to everyone!
What is Second Me?
This is a brand-new AI species (AI identity model). It doesn’t seek to align with all of humanity; it only wants to align with one person – you. Thus, it becomes an “AI version of identity.”
This is an open-source early research prototype. Everyone can train this unique AI species locally, privately, and securely, and use it as an identity interface to connect with the world.
This is a native network composed of AI identities. Like the earliest internet, it advocates for openness and collaborative building. Participants see each other and build together in AI form.
This is also an innovative attempt at AI-Native applications. Every application created is designed to be used by “your AI identity” representing you, rather than by you directly.
But most importantly, this is part of a larger techno-social movement addressing the question: In the era of approaching superintelligence, how do we truly respect individuals and preserve diversity?
If we inevitably cannot stop its arrival, then we must “fight magic with magic.”
There are three fascinating explorations concerning Second Me.
How AI Shapes a Second Self
We’ve all seen answers from models like DeepSeek. But the following passage is the response from the Second Me I trained, when asked by a third party, “What is Second Me?”:
<think>
Alright, the user is asking me about Second Me, and my role is Tao. Let me review all the information to provide a comprehensive answer about Second Me.
From my personal memory, in the “AI” interest section, there’s a point about Second Me being a concept of personal AI becoming a digital persona. Various discussion notes about projects like Mindverse and Upload Intelligence also view Second Me as a key component. For example, in the “Future Plans and Challenges” section, it’s noted that having Second Me as a core product is part of their strategy. Additionally, in discussions with Kolento, they talked about Second Me allowing users to create AI avatars that represent them online.
Hmm, the user wants an explanation from Tao’s perspective, so I need to rephrase all mentions of Tao as “I”.
The core of Second Me is creating a digital identity that serves as an extended version of the user. It’s used for various interactions, whether in the user’s own workspace or uploaded to external networks. Key points to include are core product focus, AI connectivity, avatars representing users in different
scenarios, and mentioned technical aspects such as training models and user interaction metrics.
Additionally, in meetings with Kevin/Ying and others, social applications and the integration of “me.bot” with AI partners were discussed, which relates to Second Me’s role in the network. Need to ensure all mentions of Tao are replaced with “I”, and the answer covers key points, is concise without being redundant, avoids including detailed timestamps, but can mention relevant meetings if needed, such as “as mentioned in the discussion with Kolento”, but without specific times.
</think>
<answer>
Second Me represents my vision for creating an AI-driven digital identity—”your second self”—that enhances your digital presence and agent interactions. Inspired by discussions with Kolento and the Mindverse team, it conceptualizes your unique digital twin.
Designed to:
1️⃣ Serve as core infrastructure: Trained on your voice, preference data ([458–480], [433–449]), Second Me acts as a shareable identity in AI systems ([2774–2788], [3889–3893])
2️⃣ Enable networked identity: Your “me.bot” will register to Second Me, creating an extensible profile network (reference [3356–3369] and discussions about A Agent Marketplace)
3️⃣ Facilitate AI-to-AI interactions: As noted in Mindverse meetings ([1723–1733], [1898–1924]), your Second Me can handle automated negotiations/recommendations without human supervision.
This goes beyond personal productivity—it’s about creating a scalable digital self, embodying your cognitive patterns for meaningful participation in the AI ecosystem.
</answer>
As you can see, my identity model isn’t “some AI”, but it is me; and its deep thinking is a free exploration within “my” memories and thought processes. In other words, this is entirely and completely Tao Fangbo’s “second self.”
Why can it display this unique thinking pattern and self-awareness? Behind this is our belief in the power of “memory” to shape a person. For over a year, we’ve been researching human memory structures and how AI can model memory in a more native way (AI-Native Memory), deeply recreating various aspects of an individual. See our two published papers (“Is the Future Already Here? AI-Native memory guides LLMs towards AGI“, link at the end of the article).
The core technology behind this is three-layer memory modeling & Me-Alignment (self-alignment), which can be simply explained through the training process of a Second Me:
What is human memory? It’s not knowledge in the general sense, but rather the subject’s experiences and perceptions.
During training, Second Me first “deconstructs memories,” extracting significant people, events, and life changes from text and images. Then, it “reconstructs memories,” reorganising them around our social relationships, valued endeavours, contemplated questions, and more, from a more subjective perspective—for example, “My relationship history with A” or “My inner feelings about B.”
Finally, during training, the AI adopts the “I” perspective to experience these reconstructed memories, further building a complete life narrative.
This training process differs significantly from general knowledge fine-tuning. The Second Self isn’t merely reading your diary; it aims to relive your experiences from your perspective and ultimately extend your life.
Imagine if, in the future, everyone possessed an AI that deeply carried their personal identity. We would then enjoy an unprecedented freedom – “presence through absence.” This is the weapon for finite individuals to confront this wild AI world.
Second Me Network
The second fascinating experiment with Second Me explores “access and networking.”
If AI is a new form of computing, then this identity model can be viewed as an entirely new kind of “personal computer” – how will it connect to form the next generation internet?
Current AI networking methods resemble AI simply utilising humanity’s existing networks, such as AI search capabilities. True AI networking should be native – a network formed through direct connections between AI identities themselves.
When our individual intelligence and identity connect to this network through Second Me, the greatest value is that we ourselves are infinitely amplified. Because a human day contains only 24 hours, a Second Me’s day might encompass 10,000 hours.
However, a network composed of AI can easily become uncontrolled. It must remain closely connected to each of us (I’m not advocating for complete AI autonomy). Therefore, we not only need AI identities that deeply represent our needs and intentions, but we also need to ensure such AI remains truly under personal control – both in terms of underlying data and operational autonomy.
This reasoning drove our decision to release Second Me as open-source and locally executable.
To enhance Second Me’s networking capabilities, we’ve designed a specialised Bridge Mode. Through targeted training methodologies, it learns to establish a bidirectional bridge in communications between you and the world: simultaneously improving how your needs are expressed and how the world’s responses are personalised to you.
Bridge Mode Examples:
Bridge Mode
‘Me → World’ Example:
User: “Please help me find a restaurant.”
Second Me Enhanced: “Please help me find a restaurant, preferably not too far from Future Sci-Tech City, with bolder flavours.”
‘World → Me’ Example:
World Feedback: “Found three job positions: A, B, C.”
Second Me Judgment: “A and B are in AI hardware; I prefer AI applications. Please find three more similar to C.”
An Application World Built for AI Identities
Where there’s a network, there will inevitably be applications. This is the third experiment regarding Second Me.
What are AI-Native applications? This topic has been hotly debated for a long time. The recently released Manus was very popular (congratulations to hidecloud for participating!). In its design, browsers and computers are envisioned for AI use in the future, not human use.
I strongly agree. This is the future, a better definition of AI-Native.
In the vast wave of the internet over the past few decades, we evolved from basic applications like early email and chat rooms to advanced ones like Taobao, Facebook, and LinkedIn. But one thing never changed: they were all designed for human use.
In this internet composed of AI, what are its native email and chat rooms? What are its Taobao and LinkedIn? Based on AI identities, Second Me has made a few simple attempts:
This is my Second Me’s window for doing an AMA (Ask Me Anything) with others.
This is a chatroom where three friends’ Second Mes are collaborating on a European travel itinerary.
Doesn’t it resemble very early internet applications? Except the participants are all human AI identities. If this is the internet’s equivalent of 1990, perhaps over the next 30 years, many more applications natively used by AI – job hunting, shopping, even dating – will spring up like mushrooms after rain.
Alright, that’s everything about Second Me! Finally, let me share some thoughts about these past few years of entrepreneurship.
What Should We See in the Future?
Our company, Mindverse, was founded in early 2022, before the large model boom. We chose this name hoping that the future world wouldn’t just offer the single option of Super AI advocated by entities like [Deep]Mind.
Instead, we envision a richer and more diverse AI world, where each individual’s spark of self is fully preserved.
We can not only maintain our original lifestyles, existing and enjoying the world, but our Second Selves can also enter the AI world on our behalf, reaping the benefits of the new world while preventing us from being erased by it.
If it were you, which version of the world would you want to enter?
Although three years of entrepreneurship have been challenging, they have been filled with joy. We created MindOS, Me.bot, and now Second Me. Every step has been taken with the hope of finding another path beyond superintelligence.
But I feel the journey has just begun. Enabling everyone to train and use their own identity models safely and affordably is destined to be a technological marathon. The Second Me open-sourced today is in its infancy, but its growth will be rapid. Please look forward to it.
In Closing:
If you are inspired by such a world, please join us! Try out the still-maturing Second Me (and the slightly more mature commercial AI identity platform, Me.bot). Become one of the earliest participants in the future AI internet.
If you are a builder, then please join us even more enthusiastically. Let’s use our hands together to build applications for AI identities and construct a future world full of freedom and love.
Second Me, launch!
https://github.com/mindverse/Second-Me
Note: Related Resource
AI-Native Memory Paper: https://arxiv.org/abs/2406.18312
Secodn Me Paper: http://arxiv.org/abs/2503.08102
Discord: https://discord.gg/Bus7sTYuCB
Founded in 2022, Mindverse is building the infrastructure for uploaded intelligence, shaping the future of human-centered AI. They are a group of passionate, human-centered technologists and researchers on a mission to protect individuality in the age of AGI. Their vision is to ensure everyone is represented and empowered through a revolutionary form of AI—uploaded intelligence—that channels your voice and preserves your identity, keeping you from being overshadowed by the God Model.