🔥 Big Story

Apple to Revamp Siri as Full AI Chatbot – Powered by Google Gemini

Apple announced plans to transform Siri into the company's first full AI chatbot, codenamed "Campos." In a surprise partnership on January 12, Apple revealed a multi-year contract with Google to use Gemini AI models to power Apple Foundation Models. This makes Google's Gemini the backbone of the new Siri experience.

Why it matters: After years of playing catch-up, Apple is betting big on AI—but it's doing so by partnering with a rival rather than building entirely in-house.

⚡ Quick Updates

DeepSeek V4 coming mid-February

Chinese AI startup DeepSeek is preparing to release its next flagship model around the Lunar New Year, with references to "MODEL1" already appearing in their GitHub repos.

Google launches "Personal Intelligence" in Gemini

A new beta feature lets Gemini reason across your Gmail, Photos, Search, and YouTube history for proactive, personalized responses. Rolling out to AI Pro and Ultra subscribers in the U.S.

OpenAI & SoftBank invest $1B in Stargate data centers

The joint $500M each investment in SB Energy will power a 1.2 gigawatt data center site in Texas, advancing the $500B Stargate commitment.

Microsoft launches Copilot Checkout

New agentic AI retail capabilities let shoppers complete purchases directly within Copilot. Partners include PayPal, Shopify, Stripe, and retailers like Urban Outfitters and Etsy.

📄 Top Research Papers

LLM-in-Sandbox

A novel approach enabling LLMs to explore within a code sandbox to unlock general agentic intelligence. Trains models for autonomous tool use across math, physics, chemistry, and biomedicine without task-specific training.

Possible Impact: Could significantly advance agentic AI systems by providing a new paradigm for training LLMs to interact with computer environments autonomously.

TTT-Discover: Learning to Discover at Test Time

Performs reinforcement learning at test time, allowing LLMs to continue training on the specific test problem. Achieves SOTA on Erdős' minimum overlap problem and 2x faster GPU kernels than prior art.

Possible Impact: Could revolutionize AI-assisted scientific discovery by enabling models to adapt and optimize specifically for novel problems at inference time.

Provable Robustness in MLLMs via Feature Space Smoothing

A plug-and-play module that reduces adversarial attack success rate from ~90% to ~1% across diverse multimodal LLMs—essential for safety-critical deployments.

Possible Impact: The provable robustness framework could become a standard component in deploying MLLMs in safety-critical applications.

📦 Top GitHub Repos

💻 ml-engineering | ⭐ 16.5k
The definitive open-source book for training LLMs at scale. Covers GPU debugging, distributed training, PyTorch, and MLOps.

💻 transformerlab-app | ⭐ 4.7k
Open-source ML research platform for experimenting with transformer models locally or in the cloud.

💻 fairlearn | ⭐ 2.2k
Python library for assessing and improving ML model fairness. Essential for responsible AI development.

💻 bugbug | ⭐ 557
Mozilla's ML platform that predicts bug classifications and test failures—practical AI for software engineering.

💻 aiops-modules | ⭐ 101
AWS's reusable IaC modules for SageMaker, Bedrock, and MLflow integration.

🛠️ Top AI Products

BrainLoom
Turn PDFs into flashcards instantly. A local-first Learning OS with an Infinite Canvas and "Smart Paste" for visual studying.

RunStack
Visual editor for designing LLM workflows. Build, test, and deploy prompt chains as production-ready API endpoints.

CIE - Code Intelligence Engine
Deep code intelligence for AI coding assistants like Claude Code and Cursor. 100% local, free Sourcegraph alternative.

🐦 Top Tweets

@vercel
"Skills.sh is an open ecosystem for finding and sharing agent skills."

@deedydas
"Anthropic just dropped the best blog on how to do software engg interviews in an AI world."

@sama
"We have a lot of exciting launches related to Codex coming over the next month"

@AnthropicAI
"We’re publishing a new constitution for Claude.

The constitution is a detailed description of our vision for Claude’s behavior and values"

Until next week, Stay Curious and Keep building!!
Brain Pulse

Keep Reading

No posts found