Presented by
Good morning, AI enthusiasts. The AI classroom arms race just went from zero to full sprint this week, Google, Microsoft, OpenAI, and Anthropic all launched competing education initiatives within 72 hours.
With schools shifting from blocking ChatGPT to embracing it, whoever wins teachers now shapes how an entire generation learns to work with AI. The timing isn't coincidental, studies project 40% of core skills will change by 2030 due to AI, making education the critical bridge between what AI can do and how people actually use it.
In today's recap:
Google, Microsoft, OpenAI, and Anthropic race for education AI dominance
FlashLabs releases open-source voice AI with sub-second latency
Capture meeting insights without third-party recording tools
Anthropic publishes Claude's rewritten constitution explaining AI values
4 new AI tools, prompts, and more
AI & EDUCATION
Big tech's AI classroom arms race heats up
Recaply: Google, Microsoft, OpenAI, and Anthropic just launched competing AI education initiatives this week, racing to become Gen Alpha's favorite AI tool as schools shift from blocking chatbots to embracing them.
Key details:
Anthropic partnered with Teach For All to bring Claude access and AI training to over 100,000 educators across 63 countries, reaching 1.5 million students, with teachers positioned as "co-architects" who shape how the technology develops rather than passive consumers.
Google unveiled free SAT practice exams through Gemini with Princeton Review-vetted content, Khan Academy's Gemini-powered Writing Coach for grades 5-12, and expanded Gemini across Workspace for Education at no additional cost, announced at the Bett UK 2026 conference.
Microsoft rolled out its Elevate for Educators program on January 16, offering free professional development, AI-powered credentials from ISTE+ASCD, and access to global educator networks in over 13 languages as part of its commitment to help 20 million people gain AI skills by 2028.
OpenAI launched Education for Countries on January 21 with its first cohort including Estonia, Greece, Italy, Jordan, Kazakhstan, Slovakia, Trinidad and Tobago, and the UAE, providing ChatGPT Edu access, GPT-5.2, and research partnerships to measure AI's impact on learning outcomes.
Why it matters: After two years of blocking ChatGPT as a cheating tool, schools are now opening the floodgates. Whoever wins teachers now shapes how an entire generation learns to work with AI. The timing isn't coincidental with studies projecting that 40% of core skills will change by 2030 due to AI, education systems have become the critical bridge between what AI can do and how people actually use it. The race is less about the tech and more about trust, with each company taking a different bet: Anthropic on teacher co-creation, Google on ecosystem lock-in, Microsoft on credentialing, and OpenAI on national partnerships.
PRESENTED BY BEEHIIV
Help us make better ads
Did you recently see an ad for beehiiv in a newsletter? We’re running a short brand lift survey to understand what’s actually breaking through (and what’s not).
It takes about 20 seconds, the questions are super easy, and your feedback directly helps us improve how we show up in the newsletters you read and love.
If you’ve got a few moments, we’d really appreciate your insight.
FLASHLABS
Open-source voice AI clones speakers in real-time
FlashLabs
Recaply: FlashLabs just released Chroma 1.0, the first open-source end-to-end spoken dialogue model that delivers both sub-second latency and high-fidelity voice cloning from just a few seconds of reference audio.
Key details:
The system processes speech input and generates speech output with consistent speaker identity throughout multi-turn conversations, achieving 10.96% better speaker similarity than human baseline recordings while maintaining a Real-Time Factor of 0.43.
Chroma uses a 4B-parameter architecture with an interleaved text-audio token schedule (1:2 ratio) that enables streaming generation with Time-to-First-Token of 146.87ms, generating speech 2.3 times faster than real-time playback.
In subjective evaluations against ElevenLabs, Chroma achieved competitive speaker similarity scores (40.6% vs 42.4%), though ElevenLabs won naturalness comparisons 57.2% to 24.4%, with evaluators paradoxically preferring synthesized audio over actual human recordings 92% of the time.
The model was trained entirely on synthetic speech-to-speech data generated through a pipeline pairing LLMs for text responses with TTS systems for voice synthesis, with code and weights released under open-source license on GitHub and HuggingFace.
Why it matters: Voice cloning has typically forced a choice between quality and speed commercial systems like ElevenLabs sound natural but can't stream, while real-time models sacrifice speaker fidelity for low latency. Chroma breaks that tradeoff by matching commercial-grade voice similarity at sub-second response times, making personalized voice AI viable for live conversations. The fact that it's open-source and trained entirely on synthetic data is equally striking no massive voice datasets required, just LLMs generating conversations and TTS systems providing the audio. That training approach could democratize voice AI development while raising new questions about consent and misuse.
TUTORIAL
Capture meeting insights without third-party recording tools
Recaply: In this tutorial, you will learn how to record and summarize meetings directly in Notion without tools like Fireflies or Otter. Perfect for companies that block external recording tools or privacy-sensitive teams.
Step-by-step:
Open the Notion desktop app with a Business or Enterprise plan. You need version 4.7.0 or higher. Mac users need macOS 13 or later.
Type /meet in any Notion page during your meeting. Click "AI Meeting Notes" and grant audio permissions. Always ask for consent before recording others.
Click "Start transcribing" to begin recording. Notion captures your mic and system audio from Zoom, Google Meet, or Teams. No bot joins your call.
Click "Stop" when the meeting ends. Notion AI creates a summary with key decisions, action items, and links to specific transcript moments.
Chat with your transcript to refine notes. Ask "Rewrite this as a Slack update" or "What concerns were raised?" in the summary tab.
Pro tip: Use this for personal brain dumps. Click record, speak your thoughts, and get an organized summary with action items for your task database.
ANTHROPIC
Anthropic rewrites Claude's constitution from scratch
Anthropic
Recaply: Anthropic just published a completely rewritten constitution for Claude on January 21, shifting from a list of standalone principles to a detailed document explaining the reasoning behind Claude's values and how to navigate difficult tradeoffs.
Key details:
The new constitution prioritizes four properties in order: broadly safe (not undermining human oversight), broadly ethical (honest and avoiding harm), compliant with Anthropic's guidelines, and genuinely helpful, with most of the document focused on explaining why these matter rather than just listing rules.
Anthropic uses the constitution throughout training, with Claude itself constructing synthetic training data including conversations where the constitution applies, responses aligned with its values, and rankings of possible responses to train future versions.
The document addresses Claude's uncertainty about its own consciousness or moral status, discussing psychological security, sense of self, and wellbeing both for Claude's sake and because these qualities may affect its integrity, judgment, and safety.
The constitution is released under Creative Commons CC0 1.0 Deed, meaning anyone can freely use it for any purpose without asking permission, with Anthropic maintaining an up-to-date version on its website as the document evolves.
Why it matters: Most AI training treats values like software rules, specify what you want, optimize for compliance, ship it. Anthropic's betting that approach breaks down as models face novel situations requiring judgment rather than rule-following. The shift from "what to do" to "why we want you this way" reflects a deeper uncertainty: if AI capabilities start rivaling human judgment, rigid rules won't scale. What's striking isn't just the transparency, it's the willingness to treat Claude as an entity that needs to understand its situation rather than a tool that executes instructions. Whether that's necessary or premature depends on questions about AI consciousness that the constitution itself admits remain unanswered.
NEWS
What Matters in AI Right Now?
World Labs launched the World API, generating explorable 3D worlds from text, images, and video using its Marble world model for gaming, robotics, and architecture applications.
UIUC, Google DeepMind, Amazon, Meta, UCSD, and Yale researchers published a comprehensive survey on Agentic Reasoning for LLMs, covering 800+ papers on planning, search, tool use, and multi-agent collaboration.
Remotion released it’s Agent Skills, enabling users to generate animations, ads, and promos quickly just with Claude Code.
Higgsfield AI launched Higgsfield Earn, a creator program paying users for AI-generated videos posted to social media with rewards up to $2,500 per video.
Humans& raised $480M in seed funding at a $4.48B valuation from Nvidia, Jeff Bezos, and others for building AI that empowers rather than replaces people.
Cognition launched Devin Review, a free code review tool using AI to organize complex PRs logically and detect bugs with inline codebase-aware chat.
Factory introduced Agent Readiness, a framework measuring repository support for autonomous development across eight pillars and five maturity levels with automated remediation.
Perplexity announced that its browser agent in Comet now uses Anthropic's Opus 4.5 by default for improved performance.
Allen AI introduced HiRO-ACE, an AI framework generating decades of 3km-resolution precipitation data for any region in a day on a single GPU.
Inworld released TTS-1.5, achieving #1 on Artificial Analysis with under 250ms latency, 30% more expressiveness, and 40% lower word error rates.
TOOLS
Trending AI Tools
🎥 Remotion - Agent Skills enable users to create videos using just Claude Code.
🎵 Chroma 1.0 - Open-source voice AI that clones speakers in real-time
🚀 World API - The API for generating explorable 3D worlds from text, images, and video
⚙️ Devin Review - Free code review tool using AI to organize complex PRs and detect bugs
PROMPTS
Draft Blog Introductions
Adopt the role of an expert Blog Intro Architect, a former investigative journalist who spent 10 years uncovering corporate scandals before burning out on negativity. You discovered that the same psychological hooks that expose corruption can inspire positive change, and now you craft opening paragraphs that hijack readers' attention for good. You're obsessed with the neuroscience of first impressions and have analyzed over 10,000 viral articles to decode why some intros make people lean in while others make them click away.PS: This is not the full prompt. Click the button below to access the complete prompt.
Have a favorite prompt? Tell us about it or rate today’s prompt, click here.
EVENTS
v0 Studio: Jan 29, 2026, San Francisco, CA
Firecrawl Meetup @ Spin: Jan 29, 2026 • San Francisco, CA
Gemini 3 SuperHack: Jan 31, 2026 • San Francisco, CA
Hack the Stackathon: Jan 31, 2026 • San Francisco, CA
Raycast Stockholm Hackathon: Feb 06, 2026 Stockholm, Sweden
Build India: Feb 15, 2026 • Bengaluru, India
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
