Welcome back! The tech industry is realizing that just throwing more data at a problem isn't enough anymore. To break past the current limits, the biggest players are rewriting the fundamental architecture of artificial brains, merging massive corporate empires, and putting relentless digital workers right on your desk.
In today’s Generative AI Newsletter:
Perplexity: Can a Mac mini replace human employees 24/7?
Macrohard: Is Elon Musk turning every Tesla into an office worker?
Scaling Laws: Will adding height to Transformers unlock AGI?
Nemotron 3 Super: NVIDIA solved the high cost of agentic reasoning?
Latest Developments
Perplexity Turns Mac Mini Into A 24/7 Personal AI Agent
Perplexity debuted a local version of its Computer AI agent system that runs on a dedicated Mac mini. This system gives the Comet assistant persistent access to local files and applications. It positions itself as a secure and controlled alternative to the viral OpenClaw platform.
The dedicated agent hardware:
Persistent Access: The agent manages local files and app sessions on a Mac mini from any location.
Enhanced Safety: Features include tracked activity and a kill switch to shut the system down instantly.
Model Orchestration: The system coordinates multiple different models at once to complete complex tasks.
Enterprise Expansion: A business version offers Slack integration and connections to over 400 different applications.
While OpenClaw achieved fame (and infamy) for its raw, sometimes "amok" autonomy, Perplexity is focusing on controlled persistence. By utilizing the Mac mini as a dedicated local server, the Comet assistant gains deep, always-on access to a user's files and applications without the latency or privacy risks of a pure cloud model.
Special highlight from our network
Curiosity about AI is growing across legal teams, yet the path from idea to automation often feels unclear.
e! by Lexemo bridges that gap. Through conversational “vibe coding,” lawyers describe their legal logic and the platform builds a structured workflow they can understand and refine.
Technology supports the process while legal expertise stays in charge
Special highlight from our network
Companies are actively using AI to automate routine tasks and boost productivity. More teams are bringing these tools into daily work, and the need for people who know how to use AI is growing fast.
This 2-day live AI Mastermind walks through building simple automations and using AI across common work tasks.
Strengthen your game with AI skills that employers notice.
March 14–15 | 10 AM–7 PM EST
Register now
Musk Revives Macrohard: A Joint xAI-Tesla AI Agent Project
Elon Musk announced a major strategic pivot for his AI ambitions, merging xAI’s long-rumored "Macrohard" project with Tesla’s new Digital Optimus AI agent. Musk claims the integrated system will be the first "real-time smart AI" capable of emulating the functions of entire companies by combining Grok's reasoning with Tesla's Full Self-Driving (FSD) video-processing logic.
Hybrid Hardware: The system runs on Tesla's $650 AI4 chip paired with xAI’s Nvidia servers for real-time processing.
FSD Technology: It processes live screen video and inputs by borrowing advanced training techniques from Tesla's self-driving software.
Staffing Turmoil: The announcement follows reports that over 20 engineers left the project and a major data initiative was paused.
Company Mergers: This move follows xAI's February merger with SpaceX, despite a recent wave of high-profile co-founder exits.
Macrohard is Musk's attempt to turn every Tesla into a high-powered office assistant. It is a vertically integrated dream that promises to merge the physical and digital worlds, but as with FSD, the timeline between "announced" and "operational" remains a massive question mark.
Researchers Propose Height to Overcome Stalling Scaling Laws
Traditional Transformers are hitting a wall, signaling that simply adding data or layers is no longer enough for true machine intelligence. A study in Popular Mechanics suggests the next breakthrough lies in height, a third dimension of intra-layer wiring that allows neurons within the same layer to communicate like human brain circuits.
Brain Logic:
Intra-Layer Height: Adds wiring between neurons in the same layer to enable complex, brain-like circuit communication.
Intuition Loops: Implements feedback loops that allow models to refine outputs over time, moving from uncertainty to insight.
Architectural Evolution: Prioritizes intelligent wiring and structured complexity over simply increasing raw data volume.
AGI Path: Aims to transcend current Transformer limits by creating AI that evolves toward higher-order understanding.
This design focuses on "intelligent wiring" rather than expanding scale. By mimicking human feedback mechanisms, this approach aims for cognitive evolution over raw volume. While not an immediate fix for AGI, it marks a fundamental pivot toward machines that think more like us.
Tool of the Day: NVIDIA Nemotron 3 Super
NVIDIA has launched Nemotron 3 Super, a 120-billion-parameter hybrid model purpose-built to solve the "context explosion" and high costs of agentic AI. By utilizing a specialized mixture-of-experts (MoE) architecture, it delivers 5x higher throughput and 2x higher accuracy than its predecessors, making it a powerhouse for autonomous workflows.
Core Functions:
Hybrid Reasoning Engine: Combines Mamba layers for 4x higher compute efficiency with transformer layers for advanced reasoning.
Latent MoE & Multi-Token Prediction: Activates four expert specialists for the cost of one and predicts multiple future words simultaneously to achieve 3x faster inference.
Massive 1M Token Context: A massive context window allows agents to hold entire codebases or thousands of pages of reports in memory, preventing "goal drift" during long tasks.
Blackwell Optimization: Runs in NVFP4 precision on NVIDIA Blackwell, pushing inference speeds up to 4x faster than previous-generation Hopper chips.
Try this yourself:
You can access Nemotron 3 Super today via build.nvidia.com, Perplexity, OpenRouter, or Hugging Face. Developers can also deploy it as an NVIDIA NIM microservice to build high-accuracy software development or research agents.
