Find out why 100K+ engineers read The Code twice a week

Staying behind on tech trends can be a career killer.

But let’s face it, no one has hours to spare every week trying to stay updated.

That’s why over 100,000 engineers at companies like Google, Meta, and Apple read The Code twice a week.

Here’s why it works:

• No fluff, just signal – Learn the most important tech news delivered in just two short emails.

• Supercharge your skills – Get access to top research papers and resources that give you an edge in the industry.

• See the future first – Discover what’s next before it hits the mainstream, so you can lead, not follow.

Join 100,000+ engineers who read The Code to stay ahead of the curve.

Today:

• Anthropic Supercharges Claude for Scientific Discovery

• Periodic Labs Raises $300M to Turn Failed Experiments Into AI Breakthroughs

• Deepseek OCR Compresses Text 10× for Massive AI Memory Gains

• Anthropic Launches Claude Code for Web

• OpenEvidence Hits $6B Valuation as Medical AI Soars

Anthropic Supercharges Claude for Scientific Discovery

Anthropic is upgrading Claude to assist scientists across the entire research process, not just isolated tasks. New tools, connectors, and AI “skills” aim to make Claude a reliable lab assistant for experiments, data analysis, regulatory paperwork, and discovery—helping speed up science from idea to impact.

KEY POINTS

• Full-Workflow AI Support for Scientists
  Claude now helps from early research to protocol generation, data analysis, and compliance—replacing piecemeal tasks with full research assistance.

• New Scientific Connectors & Skills
  Claude links with platforms like PubMed, Benchling, BioRender, and 10x Genomics. New “Agent Skills” allow consistent lab tasks like RNA data quality control.

• AI for Science Program & Real-World Adoption
  Anthropic partners with scientists and firms (like Deloitte, AWS) and gives free API credits to researchers tackling big problems.

Why it matters
Scientific research is slow, expensive, and often stuck in paperwork. Claude could make it faster and easier for scientists to discover new cures, test ideas, and stay organized. By helping with everything from writing study plans to analyzing gene data, AI becomes a powerful lab partner—not just a chatbot.

Periodic Labs Raises $300M to Turn Failed Experiments Into AI Breakthroughs

Two top researchers from OpenAI and Google Brain launched Periodic Labs with $300M to automate scientific discovery. Their goal: use AI, simulations, and robotics to invent new materials, starting with superconductors. Even failed experiments will feed their AI—flipping the traditional science model upside down.

KEY POINTS

• $300M for AI-Powered Lab
  Periodic Labs was founded by Liam Fedus (OpenAI) and Ekin Dogus Cubuk (Google Brain) to automate materials science using AI, robotics, and simulations.

• Scientific Method Reinvented
  Their system combines language models, physical experiments, and real-time feedback—turning every outcome, even failure, into valuable data.

• Star-Studded Backing
  Investors include Felicis, a16z, NVIDIA, Accel, and angels like Jeff Bezos and Eric Schmidt. The startup has already hired top-tier AI and science talent.

Why it matters
This is a big step toward letting AI do science, not just write about it. If successful, Periodic Labs could speed up discoveries that usually take decades—like finding better superconductors, which could lead to cheaper, more efficient energy systems. It’s a bold bet on AI’s power to shape the physical world.

Deepseek OCR Compresses Text 10× for Massive AI Memory Gains

Deepseek’s new OCR system turns image-based documents into ultra-compressed text that AI can understand using far fewer computing resources. It retains nearly all original information while shrinking memory use—making it ideal for long documents, chatbot memory, and creating huge datasets for training language models.

KEY POINTS

• Extreme Compression with High Accuracy
  Deepseek OCR compresses image-based text by up to 10× while preserving 97% of the original data, making it efficient for processing long documents or dense files like financial charts.

• High-Throughput AI Infrastructure
  With 160 GPUs across 20 servers, Deepseek can process 33 million pages per day—ideal for training large AI models or managing chatbot memory efficiently.

• Built on Meta and OpenAI Tech
  The system fuses Meta's SAM (for image segmentation), OpenAI's CLIP (image-text link), and Deepseek’s own models to create fast, multilingual document understanding in over 100 languages.

Why it matters
This tech could change how AIs read and remember. By compressing documents without losing much meaning, Deepseek’s OCR lets models handle longer conversations, richer documents, and massive training data without blowing past compute limits. It’s like giving AI a photographic memory—efficient and scalable.

🧠RESEARCH

A Theoretical Study on Bridging Internal Probability and Self-Consistency for LLM Reasoning

Researchers proposed a new method to boost how AI models reason. They showed that common techniques like generating multiple answers have flaws—some guesses are just bad. Their solution, called RPC, filters out weak answers and combines two strengths into one faster, cheaper system that still performs very well.

AI for Service: Proactive Assistance with AI Glasses

AI4Service introduces Alpha-Service, a smart assistant framework for AI glasses that helps users proactively. Instead of waiting for commands, it detects when and how to assist using real-time video, memory, and multiple specialized AI agents. Demos include helping with card games, museum tours, and shopping—no need to ask, it just helps.

From Pixels to Words -- Towards Native Vision-Language Primitives at Scale

NEO is a new kind of AI that learns to understand images and text together from scratch, without relying on separate vision and language modules. Despite using less training data, it performs competitively. The model bridges pixels and words inside one system, aiming to make vision-language AI simpler, scalable, and more accessible.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

Luma Dream Machine – AI Video Generator - video‑generation suite for the web and iOS

Mojo AI Reveals – AI Product Video Generator - AI tool for creating videos and logo animations from a single still image

Dante AI – AI Chatbot Builder - no‑code builder for AI agents

📲SOCIAL MEDIA

🗞️MORE NEWS

• Anthropic launched Claude Code on the web, letting developers assign coding tasks from their browser. It supports mobile, GitHub integration, real-time feedback, and runs securely in isolated environments, helping users fix bugs and speed up development.

• OpenEvidence, a medical AI tool likened to ChatGPT for doctors, raised $200M at a $6B valuation. Just three months after raising $210M, it doubled its valuation. Trained on trusted medical journals, OpenEvidence offers free, ad-supported access to verified clinicians and now handles 15M monthly consultations. Google Ventures led the round.

• IBM is integrating Groq’s lightning-fast AI into its watsonx platform. This gives enterprise users faster, cheaper AI tools via GroqCloud. Use cases include healthcare and retail. Groq’s chip outpaces GPUs, and Red Hat tech and IBM models will also run on the platform.

• Adobe launches AI Foundry to deeply customize Firefly for enterprise brands. Instead of simple tweaks, Adobe will rebuild Firefly models with each company’s brand tone, visuals, and IP. The result: multimodal, enterprise-specific models, starting with clients like Home Depot and Disney Imagineering.

• Meta AI app surges after launching Vibes video feed. Daily users jumped to 2.7M and downloads hit 300K/day, likely driven by the new short-form AI video feed and Sora’s invite-only model, which may have pushed curious users to Meta instead.

• NVIDIA and Google Cloud launch G4 VMs with RTX PRO 6000 Blackwell GPUs to accelerate enterprise AI. Now available, these powerful virtual machines support agentic AI, robotics, simulation, and high-end visual computing. Tools like Omniverse and Isaac Sim unlock digital twin and robotics workflows at industrial scale.

• Wikipedia traffic drops as AI and social media bypass links. Page views are down 8% year-over-year, with users now getting content directly from AI tools and social platforms. Wikimedia warns this could threaten Wikipedia's future and is testing new formats to adapt.

• Meta bans all general-purpose chatbots on WhatsApp—except its own. Starting Jan 15, 2026, tools like ChatGPT and Perplexity will be blocked from WhatsApp. Only Meta AI will remain, as Meta aims to control platform use and reduce system strain.

• A new AI system by OpenVet scored 100% on the NAVLE, the licensing exam for veterinarians in the U.S. and Canada. Though not yet launched, it shows promise as a trusted reasoning engine and reflects growing adoption of AI in veterinary medicine.