Meta Moves, Models & Machines: The Rise of Physical Intelligence
16th June 2025 | Superintelligence Newsletter
Hey Superintelligence Fam 👋
Meta’s ambitious leap into superintelligence, led by Scale AI’s Alexandr Wang, isn't just about smarter models—it’s about grounding AI in the physical world. Enter V-JEPA 2, Meta’s video-trained world model.
This powerful system marks a turning point in robotics, blending predictive intelligence with real-world planning. Physical intelligence is no longer sci-fi—it's the next frontier in AI-human collaboration and autonomous action.
Struggling with late payments?
Learn proven strategies to speed up cash flow and ensure you get paid on time. Download this free guide and take control of your revenue today!
Meta Forms Superintelligence Lab, Taps Scale AI’s Alexandr Wang : Meta commits $14.3 B for a 49% stake in Scale AI, tapping CEO Alexandr Wang to spearhead its new “superintelligence” lab, marking a moonshot push towards AGI leadership.
AMD Unveils Helios AI Server with MI350 and MI400 Chips : At its “Advancing AI” conference, AMD unveiled MI350 and MI400 series GPUs, previewing its Helios rack-scale AI server with 72 chips, and announced that OpenAI will adopt its latest processors.
Meta Introduces V-JEPA 2 Video-Based World Model : Meta unveils V-JEPA 2, a video-trained world model enhancing AI agents’ physical reasoning, and rolls out three new benchmarks to evaluate models’ visual prediction and planning capabilities.
Text-to-LoRA: Instant Transformer Adaption : T2L compresses nine pre-trained LoRA adapters into a hypernetwork, generating task-specific LoRA weights in a single pass, achieving zero-shot performance matching fine-tuned baselines with minimal compute overhead.
V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning : V-JEPA 2 pre-trains a 1.2B-parameter JEPA model on 1M+ hours of video, hitting 77.3% top-1 on Something-Something v2 and enabling zero-shot robotic pick-and-place with 62h data.
Reinforcement Pre-Training
RPT reframes next-token prediction as RL with intrinsic verifiable rewards, scaling on unlabelled corpora to boost prediction accuracy and reasoning, with RPT-14B matching bigger models on OmniMATH.
Google AI Edge : Google AI Edge brings powerful on-device AI to mobile, web, and embedded platforms with low-code APIs, cross-platform frameworks, hardware acceleration for seamless offline intelligence.
Eleven v3 : Eleven Labs’ v3 alpha delivers human-like, multi-speaker TTS with emotional depth, inline audio tags, and global language support across 70+ languages for immersive AI audio.
o3 Pro API : OpenAI’s o3-pro API delivers unparalleled reliability and tool mastery—real-time web search, Python execution, and powerful vision reasoning—trading speed for precision at $20/$80 per million tokens.
Unlock the Power of Large Language Models
Delve into the world of AI with a free 7-day email course covering LLM fundamentals, architectures, training processes, applications, and ethical considerations.
Enroll Now to enhance your AI expertise.
Between June 9–15, 2025, BFI warned AI training on 130,000+ scripts risks the UK screen sector, US bishops urged Congress to enshrine AI ethics for human dignity, Seattle University unveiled its Ethics & Tech conference on AI governance, and a cybersecurity strategist cautioned that delaying state-level oversight imperils national security.
Struggling with late payments?
Learn proven strategies to speed up cash flow and ensure you get paid on time. Download this free guide and take control of your revenue today!
Thank you for tuning in to this week's edition of Superintelligence Newsletter! Stay connected for more groundbreaking insights and updates on the latest in AI and superintelligence.
For more in-depth articles and expert perspectives, visit our website | Have feedback? Provide feedback.
If you wish to partner with us then Explore Here
Stay curious, stay informed, and keep pushing the boundaries of what's possible!
Until Next Time!
Superintelligence Team.