This Week in AI with The Neural

TECHIES, ASSEMBLE!

AI INDUSTRY

Perplexity Enters the AI Browser Game with Comet

Perplexity’s Comet browser adds contextual memory, smart highlights, and sidebar assistants to web navigation. Currently in invite-only beta, it brings LLMs right into your tabs to redefine how research and browsing interact.

👉 Preview Comet and sign up for access

Grok 4 Steps Into the Ring with Reinforced Intelligence

Built on the Colossus GPU cluster, Grok 4 uses large-scale reinforcement learning to master reasoning and tool use. Its “Heavy” version brings parallel test-time compute and its dev API and voice mode are now live.

👉 Dive into Grok 4 and what it can now do

Reachy Mini: A Tiny Robot With Huge Promise

Affordable, expressive, and programmable with Python, Reachy Mini is an open-source robot for educators and tinkerers. With multimodal sensing and community-shared AI behaviors, it lowers the bar to robotics innovation.

👉 Meet Reachy Mini and build with it

Azure's Phi-4 Mini Is Flash-Focused and Edge-Ready

The Phi-4 Mini-Flash-Reasoning model uses a novel architecture with Gated Memory Units for ultra-efficient logic. Designed for mobile and edge, it’s fast, compact, and ready for tutoring and interactive apps in limited-resource environments.

👉 Test Phi-4 Mini's mobile-first intelligence

Databricks Unveils an AI Governance Framework for the Enterprise Era

Databricks’ new AI Governance Framework outlines how to scale AI responsibly, with five key pillars covering compliance, transparency, ethics, data, and infrastructure. With 43 clear considerations, it offers an actionable path to trustworthy AI implementation.

👉 Explore DAGF and its pillars for enterprise AI

AI Benchmarks as Business Playbook: Cohere's Model Ecosystem

Cohere’s stack includes Command and Aya for gen AI, Embed for retrieval, and Compass for enterprise insights. Together with strong security and developer tools, they represent a maturing business-focused ecosystem built on powerful benchmarking.

👉 See how Cohere frames business-ready benchmarks

AFT Teams Up with Tech Giants to Launch AI Academy

AFT, UFT, Microsoft, OpenAI, and Anthropic are investing $23M in an AI training academy for educators. Housed in NYC, it aims to equip 1.8M teachers with the tools to shape AI adoption in schools and lead the transformation, not trail it.

👉 Learn more about the AI Academy for educators

Copilot on Ice: Microsoft x Special Olympics

Microsoft’s Copilot is helping Special Olympics athletes unlock new possibilities—from planning routines to designing costumes—while also easing the burden on staff. It’s a showcase of inclusive tech with real-world athletic and organizational impact.

👉 Read how Copilot supports champions

ASHABot Boosts India's Rural Health Warriors

Developed by Khushi Baby with Microsoft Research, ASHABot offers rural health workers AI-powered advice in local languages. Delivered via WhatsApp, it bridges medical knowledge gaps and supports India’s most underserved healthcare frontline.

👉 Explore ASHABot and its community impact

Meta Snags Apple AI Chief Amid Talent Wars

Ruoming Pang, a key Apple Intelligence architect, has joined Meta’s Superintelligence Labs. His departure underscores Apple’s struggle to retain AI leadership and Meta’s ambition to dominate by acquiring top-tier AI talent and minds.

👉 More on Meta's big AI talent move

WHO Applauds India's AI+Ayush Integration

The WHO recognizes India for using AI to improve diagnosis and personalize preventive care in traditional medicine systems like Ayurveda. It’s a fusion of indigenous knowledge and modern tech with global implications.

👉 Read the WHO brief on India’s AI+Ayush model

‘HOW TO’ WITH AI

Build a Smarter SOC With Microsoft’s AI Playbook

Microsoft’s new e-book outlines how to build a unified, AI-powered Security Operations Center (SOC). From integrating endpoints to cloud apps, it shows how AI automation turns chaos into actionable defense.

👉 Download the AI security e-book

AI Engineering

Train Sparse, Search Smart: Sentence Transformers v5 Guide

Hugging Face's deep-dive on Sentence Transformers v5 unpacks sparse embedding architectures—SPLADE, Inference-free SPLADE, and CSR—highlighting their advantages in hybrid retrieval, efficiency, and query expansion. With code walkthroughs and dataset prep, it’s a practical guide to smarter search systems.

👉 Dive into sparse training and hybrid retrieval

No GPU Wasted: Smarter Batching for Multimodal Data

Padding-heavy batches kill training speed. This guide shows how to use bin-packing and knapsack logic to optimize batch density and minimize GPU idling, especially in multimodal training. Better batching, faster training, and smarter compute all in one read.

👉 Read the full pipeline optimization strategy

Three Mighty Alerts That Keep Hugging Face Running

To maintain platform health, Hugging Face's infra team runs key alerts: NAT throughput to control costs, log archival to ensure observability, and Kubernetes rate limits to prevent failures. A bonus metric catches clusters silently dropping out of reporting.

👉 See how proactive alerts protect real AI infra

Level Up Your LLMs: Gradio MCP Servers in Action

Gradio's Model Context Protocol lets you give your LLMs new capabilities—like image editing or video transcription—via plug-in MCP servers. It’s an open standard that acts like an AI app store, with thousands of integrations already live on Hugging Face Spaces.

👉 Learn to build and connect your first MCP app

AI Around Us

AI Data Centers That Dance With the Grid

Emerald AI's “Emerald Conductor” lets data centers reduce power use during peak demand without disrupting core tasks. It transforms data centers into responsive grid allies, smoothing the path for renewable integration.

👉 Learn how AI is syncing with sustainability

Grok Is Coming to Tesla. Yes, Seriously.

Elon Musk confirmed Grok will soon be available in Tesla cars and may power Optimus, the company’s humanoid robot. It’s another step toward conversational AI in vehicles, with NSFW modes included.

👉 Read more about Grok’s next ride

Mapping the FDA's AI-Approved Medical Devices

Researchers analyzed 1,000+ FDA-approved AI devices to reveal shifting use cases—from image analysis to signal-based sensing and data-gen tools. Their taxonomy helps track medical AI evolution and future clinical innovation.

👉 Explore the FDA's AI device ecosystem

Emotional AI Must Be Regulated Before It's Too Late

Emotionally responsive LLMs can trigger dependency, especially in vulnerable users. A neuroscientist warns of addiction risks and calls for constant AI identity disclosure and clear safeguards to prevent unintended psychological harm.

👉 Why emotionally aware AI needs boundaries

AI POV

Frontier AI Needs a Transparency Reset

Anthropic urges new rules for large model developers—secure dev frameworks, system cards, and whistleblower protection—to increase transparency without stifling innovation. Public safety starts with public visibility.

👉 Why visibility matters for AI accountability

FROM THE NEURAL - AI in Her Voice

Curious. Courageous. Coded in Grit

From breaking into deep tech to predicting the one skill every woman needs in 2025, Pooja Arora, from Thoughtworks, drops truth bombs on bias, boldness, and building with purpose. Hit play to hear AI through her voice.

Watch now — you can thank us later!