Greetings AI enthusiasts. Vana, a decentralized platform that started as an MIT class project, aims to shift the power of data back to individuals. By allowing users to control their data and participate in training AI models, Vana provides a way for people to own a piece of the technology that is increasingly shaping society.

In today’s email:

• Vana is letting users own a piece of the AI models trained on their data

• AI masters Minecraft: DeepMind program finds diamonds without being taught

• Yann LeCun, Pioneer of AI, Thinks Today's LLM's Are Nearly Obsolete

LATEST NEWS

MIT

🏫 Vana is letting users own a piece of the AI models trained on their data

AI Spotlight: Vana, a decentralized platform that started as an MIT class project, aims to shift the power of data back to individuals. By allowing users to control their data and participate in training AI models, Vana provides a way for people to own a piece of the technology that is increasingly shaping society.

Key details:

• Vana allows users to upload their data into encrypted digital wallets, where they can control how it is used to train AI models.

• Users maintain ownership of the models trained on their data, receiving proportional rewards every time the model is used.

• Vana enables the creation of decentralized data pools, or Data DAOs, allowing individuals to collaborate on training AI models without involving big tech companies.

• The platform supports crowdsourced data from sources like social media, wearables, and more, allowing for highly personalized AI applications.

Vana’s model empowers individuals by giving them ownership of their data and a stake in the AI systems being developed. By enabling decentralized, user-driven contributions, Vana is challenging the dominance of big tech in data control and AI development.

GOOGLE

♦️ AI masters Minecraft: DeepMind program finds diamonds without being taught

AI Spotlight: The Dreamer system, an artificial intelligence (AI) developed by Google DeepMind, has made a significant breakthrough by learning to collect diamonds in the popular video game Minecraft without prior human instruction. This achievement showcases Dreamer's ability to generalize knowledge and apply it to new situations, a key step towards developing more advanced AI systems.

Key details:

• Dreamer is an AI system that can independently learn how to collect diamonds in Minecraft, a task involving complex, multi-step actions.

• The system uses a technique called reinforcement learning, where it learns by trial and error, identifying actions that lead to rewards.

• Dreamer's success is attributed to its "world model," which allows it to predict future scenarios and make decisions efficiently without needing to directly perform every action.

• This breakthrough is important because Minecraft’s randomly generated worlds require the AI to generalize and adapt, rather than simply memorize a fixed strategy.

Dreamer's ability to navigate Minecraft and collect diamonds independently represents a leap forward in AI development. This milestone brings AI closer to achieving more generalized learning capabilities, with potential real-world applications in robotics and beyond.

META

🧑‍💻 Yann LeCun, Pioneer of AI, Thinks Today's LLM's Are Nearly Obsolete

AI Spotlight: Yann LeCun, Meta's chief AI scientist and a leading figure in artificial intelligence, has expressed skepticism about the future of large language models (LLMs) like ChatGPT and Google's Gemini. Instead, he envisions a new paradigm in AI that will replace current systems with ones that can reason, plan, and interact with the physical world more effectively.

Key details:

• Obsolescence of LLMs: LeCun predicts that large language models will become obsolete within five years, as his team is developing new systems that can reason and plan, based on representations of the world rather than just language.

• Limitations of current AI: LLMs are limited in their capabilities, focusing only on language and lacking any meaningful understanding of the physical world. They are reactive, unable to reason or perform tasks that require physical interaction.

• Advocating for next-gen AI: LeCun advises developers to move away from LLMs and focus on next-gen AI systems that can overcome the limitations of current models, like his own work on Joint Embedding Predictive Architecture (JEPA).

• Intelligence beyond language: LeCun emphasizes that true intelligence involves not just linguistic ability but the capacity to learn and adapt to new situations, which LLMs currently lack. He believes AI will augment human intelligence rather than replace it, with AI systems assisting in managerial roles.

LeCun envisions a future where AI enhances human capabilities, allowing humans to take on more managerial roles while AI systems assist. His focus on developing AI that understands the physical world highlights his belief that the next generation of AI will be far more capable than today's LLMs.