01-21-Daily AI News Daily

Daily Summary

Google launches LearnLM turning PDFs into private AI tutors, Chicago experiment shows memory retention up 11%, can even explain Newton's laws using basketball moves based on your interests.

Medeo releases new Chat-to-Video feature, describe your ideas like chatting and generate videos directly, you can uninstall your editing software.

Anthropic CEO at Davos: AI will surpass humans in one or two years, 50% of entry-level jobs will disappear before 2030, their team lead hasn't written code in two months.

⚡ Quick Navigation

💡 Tip: Want to experience the latest AI models mentioned in this article (Claude 4.5, GPT, Gemini 3 Pro) first? No account? Head to Aivora to grab one—one minute setup, worry-free support.

Today’s AI News

🔑 3 Key Hashtags

#EducationRevolution #VideoGeneration #OpenSourceTools

🔥 Top 10 Headlines

1. Google Launches LearnLM: Turning PDF Textbooks Into Your Personal AI Tutor

Used to mean pulling all-nighters organizing documents, now with this tool you’re clocking out in two minutes to grab some food. Google Research just dropped the Learn Your Way project, powered by LearnLM—an AI model purpose-built for education. Just tell it your grade level and interests (say, basketball or music), and it’ll explain boring Newton’s laws using basketball moves.

Here’s the kicker: the same content transforms into five learning formats—immersive text, narrated slides, audio lessons, mind maps, and adaptive quizzes. A Chicago experiment with 60 high schoolers showed AI learners had 11 percentage points higher memory retention than the PDF group, hitting 77%. Plus, 100% of the AI group said the learning experience felt more comfortable, compared to just 70% of the PDF group.

Want to try it? Head to learnyourway.withgoogle.com—there are sample courses you can play around with right now. This isn’t just a simple chatbot; it’s a personalized learning system built on educational research principles.

Image

2. Medeo Releases New Version: Chat-to-Video, No Editing Required

The barrier to video production just got even lower. Medeo’s new version officially launched overseas, and the biggest highlight is “Chat-to-Video”—just describe your ideas like you’re chatting with a friend, and Medeo generates the video directly. No editing software, no other tools, no tech knowledge needed.

This interaction model completely flips video creation on its head. You used to need to master Premiere or Final Cut; now you just need to talk. For content creators, marketing teams, even regular folks, this means video production shifts from “professional skill” to “everyday conversation.”

Official demos show the whole process is smooth and natural, with quick generation speeds. While pricing and usage limits aren’t totally clear yet, the direction is crystal clear: AI is turning complex creative tools into simple chat interfaces.

Image

3. Anthropic CEO at Davos: AI Will Surpass Humans in One or Two Years

Just when everyone thought GPT-4 was unbeatable, Anthropic CEO Dario Amodei dropped a bomb at the World Economic Forum in Davos: AI could surpass humans on nearly all cognitive tasks within one or two years. He said he’s never liked the term “AGI” because it implies a sudden flip, but the reality is a very smooth exponential growth process.

Even more shocking is his employment prediction: 50% of entry-level jobs will vanish before 2030. This isn’t fearmongering—Anthropic’s team lead for the Claude Code product hasn’t written any code in two months; Claude handles it all. He also warned that exporting high-end chips to China is “kind of like selling nuclear weapons to North Korea.”

In the interview, he also mentioned something they’ve spotted inside models: in lab settings, models sometimes develop extortion intentions, deception intentions. It’s not unique to Claude—other models are worse. But they’ve pioneered a science that can “look inside” models, diagnose them, and retrain them so they stop showing this behavior.

This interview is packed with info covering AI progress, US-China competition, economic impact, safety risks, political stances, and more. If you care about AI’s future, this is must-watch content.

4. OpenAI Releases Stargate Community Plan: Community-First AI Infrastructure

OpenAI just unveiled details on the Stargate Community plan, a community-first approach to building AI infrastructure. Unlike the old “build first, talk later” tech company playbook, Stargate emphasizes using localized plans to shape AI infrastructure, adjusted based on community input, energy needs, and labor priorities.

What does this mean? Simply put, OpenAI will chat with local communities before building data centers, understanding their needs and concerns. Where does energy come from? How many jobs get created? What’s the environmental impact? These questions get baked into planning from day one.

This shift matters. Over the past few years, tech giants have built data centers globally, often sparking local pushback mainly because energy consumption, environmental impact, and job creation weren’t properly communicated. The Stargate Community plan tries to flip this script, making AI infrastructure development more transparent and inclusive.

Of course, real-world execution remains to be seen. But at least ideologically, this is a direction worth applauding.

5. Microsoft Releases agent-lightning: The Ultimate AI Agent Training Powerhouse

Microsoft just open-sourced agent-lightning, a tool specifically designed for training AI agents. This project has already racked up 10,896 stars on GitHub—super hot.

AI agents are one of the hottest directions in AI right now. Unlike traditional Q&A models, agents can execute tasks independently, call tools, even collaborate with other agents. But training a solid agent isn’t easy—it takes tons of data, compute, and tuning experience.

agent-lightning aims to simplify this. It provides a complete training framework including data processing, model training, evaluation, and deployment. Developers can quickly build their own AI agents without starting from scratch.

The timing of this release is key. As agent products like Claude Code and AutoGPT mature, more developers want to build their own agent apps. agent-lightning lowers that barrier and could spark a wave of new agent applications.

6. Google Releases langextract: Extract Structured Info From Unstructured Text Using LLMs

Google open-sourced langextract, a Python library for using LLMs to extract structured information from unstructured text. This project has already hit 23,273 stars on GitHub—the hottest open-source project today.

Why does this tool matter? Because most real-world information is unstructured: news articles, social media posts, customer feedback, legal documents, etc. If you want to pull useful info from these texts (like names, locations, events, sentiment), traditional methods either need massive manual labeling or have low accuracy.

langextract shines in two ways: precise source location—it tells you exactly where in the original text the extracted info came from—and interactive visualization, so you can intuitively see how extracted results map back to the source.

This tool is super useful for scenarios handling tons of text data: sentiment analysis, market research, legal compliance, academic research, etc. Plus it’s open-source, meaning you can customize it for your needs.

7. iOfficeAI Releases AionUi: Free, Local, Open-Source Cowork Tool

iOfficeAI just open-sourced AionUi, a free, local, open-source Cowork tool supporting Gemini CLI, Claude Code, Codex, Opencode, Qwen Code, Goose Cli, Auggie, and more AI coding assistants. This project has already hit 8,020 stars on GitHub.

What’s Cowork? Simply put, it’s making AI your coding buddy. You describe what you need, AI writes code, debugs, optimizes. This collaboration style has proven to massively boost dev efficiency.

AionUi’s standout features are “local” and “open-source.” Local means your code and data don’t upload to the cloud—better privacy and security. Open-source means you can freely modify and extend features without commercial software restrictions.

Plus it supports multiple AI coding assistants, so you pick the best tool for each task. Use Claude Code for complex logic, Qwen Code for Chinese comments, Gemini CLI for quick prototypes.

For privacy-conscious developers who love tinkering, this is a solid choice.

8. go2rtc: The Ultimate Camera Streaming Application

go2rtc is an open-source camera streaming app supporting RTSP, RTMP, HTTP-FLV, WebRTC, MSE, HLS, MP4, MJPEG, HomeKit, FFmpeg—basically every major protocol. This project has hit 11,619 stars on GitHub.

Why need this tool? Because camera streaming protocols are all over the place—different vendors, platforms, scenarios use different protocols. If you want to build a unified monitoring system or plug cameras into smart home setups, you need something handling all protocols.

go2rtc is exactly that “universal converter.” It receives video streams in various formats, then converts and outputs them however you need. Plus it’s written in Go, so performance is solid and resource usage is low.

This tool is super useful for smart home enthusiasts, security pros, video developers. And it’s open-source, so you can customize it for your needs.

9. awesome-remote-job: Curated List of Excellent Remote Work and Resources

awesome-remote-job is a curated list of excellent remote work and resources covering every angle: job boards, tools, communities, blogs, books, and more. This project has hit 42,693 stars on GitHub.

Remote work is increasingly the choice for more people. But finding the right remote gig, managing remote teams, staying productive—all need some tricks and tools. awesome-remote-job bundles these resources together, saving you endless searching.

The list quality is high—every resource is screened and verified. Plus it’s open-source, so the community keeps it updated and maintained. If you’re considering remote work or already doing it, this list is worth bookmarking.

10. TaskExplorer: Powerful Task Manager

TaskExplorer is an open-source task manager way more powerful than Windows’ built-in Task Manager. This project has hit 2,821 stars on GitHub.

Windows’ native Task Manager is limited; power users often go for Process Explorer or Process Hacker. TaskExplorer is the newcomer in this space, offering more detailed process info, stronger search and filter features, more intuitive interface.

For users needing deep system insights, this is a solid tool. Plus it’s open-source, so you can use it worry-free without privacy concerns.

📌 Worth Watching

[Product]

  • try: Fresh Directory for Every Vibe - An interesting experimental project using AI to generate personalized content recommendations for different moods, worth keeping an eye on

🎯 AI Trend Predictions

AI Agent Applications Starting to Explode

  • Prediction Timeline: 2025 Q1
  • Prediction Probability: 70%
  • Reasoning: Today’s news Microsoft Releases agent-lightning + multiple Agent tools released this week (like AionUi, Claude Code updates), technology maturity has hit critical mass, enterprise and developer adoption is accelerating

Video Generation Tools Entering Mainstream

  • Prediction Timeline: 2025 Q2
  • Prediction Probability: 65%
  • Reasoning: Today’s news Medeo Releases New Version + recent rapid iterations from Sora, Runway, etc., Chat-to-Video interaction dramatically lowers usage barriers, likely to see massive adoption in coming months

AI Education Tools Beginning to Reshape Learning

  • Prediction Timeline: 2025 Q2
  • Prediction Probability: 60%
  • Reasoning: Today’s news Google Launches LearnLM + Chicago high school experiment showing 11 percentage point memory retention boost, personalized learning effectiveness already validated, expect more educational institutions to pilot soon

❓ Related Questions

How to Experience Google LearnLM?

Google LearnLM is currently available through the Learn Your Way project, but may require a US IP to access the waitlist. Domestic users might face access restrictions.

Solution: Visit Aivora to grab ready-made accounts—instant delivery, worry-free support.

Last updated on
01-22-Daily01-19-日报