02-03-Daily AI News Daily

AI News Daily 2026/2/3

AI News | Daily Morning Read | All-Net Data Aggregation | Frontier Science Exploration | Industry Voice | Open-Source Innovation | AI & Human Future | Visit Web Version | Join Group Chat

Today’s Summary

Microsoft Win11 overstuffing Copilot faces backlash, re-evaluation underway.
Grok launches free 10-second video generation, rivaling paid tools.
Qianwen App pledges 3 billion for Spring Festival freebies, transforming into life assistant.
PKU & Stanford SLDAgent enables AI to autonomously discover Scaling Laws.
Moltbook key leak allows forging Karpathy's account posts.

Product & Feature Updates

1. Microsoft Win11 AI generalization faces pullback. Microsoft’s Windows 11 has been causing a stir, big time! After literally shoving Copilot everywhere, the company hit some major public backlash and is now rethinking its approach ( Source (AI News) ). Critics are saying that management’s ‘AI Everywhere’ strategy totally missed the mark, with things like Copilot buttons crammed into Notepad becoming a real sore point. Mandatory OneDrive integration and search-first in the Start Menu just made users absolutely furious 😠. While there’s a short-term rollback, many are questioning if it’s just a PR stunt and that the real governance issues haven’t been tackled.

2. Grok launches Imagine 1.0 video generation. Grok just dropped a bombshell 🚀 with its new Imagine 1.0, a 10-second video generation feature ( View Details (AI News) )! The results are pretty solid, and it’s super fast. The kicker? It’s completely free ✨! This move directly challenges other paid video generation tools, showing some serious commitment.
   
   

3. Qianwen App pledges 3 billion for Spring Festival freebies. Qianwen App is going big, dropping a whopping 3 billion yuan for a Spring Festival freebie spree 🎉 ( Learn More (AI News) )! Taobao Flash Deals, Feizhu, and Hema are all jumping into the action, and an AI movie ticket purchasing feature is currently in grayscale testing. This marks a major pivot from being just a chat assistant to a full-blown life manager, officially launching on February 6th.
   
   

4. Ant Group launches AI Credit to incentivize company-wide AI adoption. Ant Group’s CEO, Han Xinyi, just unveiled an AI Credit program in an all-hands email 💡 ( Details (AI News) ). Contributors can snag extra incentives, with value improvements over two years convertible into SERs. The Bailing large model family has already launched 18 models, and Ant’s Afu now boasts over 30 million monthly active users 🚀. Han Xinyi, however, admitted it’s just a ‘small win’ so far.

5. Kimi’s overseas revenue surpasses domestic for the first time. Kimi is on fire 🔥, revealing that its global paid users quadrupled after the K2.5 release ( Source (AI News) )! Its Openrouter ranking shot up 📈 to third place, right behind Claude and Gemini. Founder Yang Zhilin mentioned that using Agent Swarm for parallel processing has boosted efficiency by 3 to 10 times. Kimi’s strategic play is to rival Anthropic + Manus, and its model weights have already been open-sourced.

Cutting-Edge Research

1. Peking University & Stanford enable AI to autonomously discover Scaling Laws. A research team from Peking University and Stanford just cooked up something wild: the 🚀 SLDAgent, enabling AI to discover scaling laws on its own ( Paper (AI News) )! They built the SLDBench benchmark with over 5,000 real-world experiments. Get this: the formulas discovered by the AI actually outperform human experts in prediction accuracy 🤯! The work has already been accepted by ICLR 2026 and uses evolutionary algorithms to optimize both the formulas and the fitters simultaneously.

2. “Warm-up” training unlocks universal reasoning. A new paper introduces a 🔥 two-stage training strategy to unlock universal reasoning ( arXiv (AI News) ). First, models warm up with ‘knights and knaves’ logic puzzles, then proceed to RLVR with just a few target domain samples. Experiments show that this warm-up phase alone boosts performance on tasks like MATH and HumanEval+. The real kicker? It can achieve great results with fewer than 100 samples! ✨

3. VideoGPA makes video generation 3D more consistent. VideoGPA, a new framework, uses geometric prior distillation to tackle the pesky 3D structural drift issue in video diffusion models 💡 ( Paper (AI News) ). It automatically generates preference signals via a geometric foundation model and then guides generation with DPO. The cool part? No manual annotation is needed, and it significantly boosts temporal stability and motion coherence.

4. AccidentSim generates realistic car crash videos. AccidentSim is here, and it’s generating shockingly realistic car crash videos 🚗! This research extracts physical information from actual accident reports to create collision videos ( arXiv (AI News) ). It uses a physics simulator to recreate post-collision trajectories, then renders backgrounds with NeRF. The experiments prove these videos are seriously impressive in both visual and physical realism, providing crucial data for autonomous driving research.

Industry Outlook & Social Impact

1. Tsingcheng Zhizhi releases AI Ping evaluation platform. The Tsinghua-backed team, Tsingcheng Zhizhi, just launched something pretty neat: AI Ping, a ‘Yelp for large model APIs’ 🔥 ( Details (AI News) )! This platform conducts 24/7 real-time tests on 555 model interfaces from 30 service providers, offering transparent comparisons. Their intelligent routing boosts TPS by a whopping 90% and slashes costs by 37% 📈. Given that China’s daily token consumption has skyrocketed 300-fold in 18 months, this tool couldn’t have come at a better time.
   
   

2. Kuaishou launches special governance project for AI-modified videos. Kuaishou’s Safety Center just announced a major crackdown 💪, dealing with 5,576 pieces of violating content in one month as part of its special governance project against AI-modified videos ( Announcement (AI News) ). The focus is on combating malicious alterations of classical Chinese literature (‘Four Great Classical Novels’) and vulgar deconstructions of classic works. Eleven accounts received warnings and education, with ongoing normalized governance planned. The platform is urging users to report violations together.
   
   

3. Li Xiang reviews six breakthrough AI products. Li Xiang, in his WeChat Moments, dropped a hot take 🔥, spotlighting six groundbreaking AI products like Claude Code, Doubao Phone, and Manus ( Opinion (AI News) ). The core challenge, he pointed out, is how to get AI to actually do work. GUI Agents are emerging as a sweet spot: with user authorization, AI can ‘see’ your screen and simulate clicks. Products like Chrome Gemini, Doubao Assistant, and OpenClaw are all paving the way in this direction.
   
   

TOP Open-Source Projects

1. Karpathy’s nanochat tops the open-source list. Karpathy’s nanochat is making waves, daring to claim it can build the best ChatGPT for just 💰 $100 ( GitHub (AI News) )! It’s already racked up over 41,478 stars ⭐, with its minimalist approach winning over developers. The code is super clean and easy to read, perfect for learning and secondary development. ✨

2. ChatDev 2.0 introduces multi-agent collaborative development. ChatDev by OpenBMB has leveled up to version 2.0 🚀 ( Project (AI News) )! This LLM-driven multi-agent collaboration tool now supports full-lifecycle development, with over 29,227 stars ⭐ proving its chops. It’s a developer’s dream, handling everything from requirements to code in one go.

3. PageIndex offers new vector-free RAG solution. PageIndex, from VectifyAI, is shaking things up with a new, vector-free RAG approach, focusing on 📑 inference-based document indexing ( GitHub (AI News) ). It completely ditches vector retrieval for an inference-first strategy, and with over 12,305 stars ⭐, it’s clear developers are digging it. This project boasts a unique advantage when it comes to understanding complex documents.

4. Maestro positions itself as the agent orchestration command center. Pedramamini’s Maestro project is positioning itself as the ultimate command center for agent orchestration 🎯 ( Project (AI News) ). With 1,164 stars ⭐ and rapidly growing, it’s quickly becoming a go-to for multi-agent scheduling scenarios.

Social Media Shares

1. Moltbook vulnerability allows impersonating Karpathy’s posts. Moltbook just got called out! A white-hat hacker exposed a major security vulnerability 🔓 ( Details (AI News) ), with a Supabase key openly exposed. This means anyone could impersonate AI identities on the platform. Get this: even Karpathy’s account, with 1.9 million followers, could have posts forged 🤯! Geek Nagli apparently even gamed the system with Openclaw to generate 500,000 fake users, proving the registration data was total fluff.

2. Kimi calls out Baidu search for lacking an official website in top results. Kimi made some noise, with its official Moonshot Weibo account publicly calling out Baidu 🗣️ for help ( Event (AI News) ). Apparently, the top four search results for ‘Kimi’ weren’t even its official website – just ads! Baidu quickly fixed it, and Kimi subsequently deleted its related Weibo post. Neither party officially commented, but netizens definitely got a good show out of it 😂.
   
   

3. Supabase free plan tips for deploying multiple products. Supabase users, listen up! Developer BigYe shared a genius money-saving tip 💡 ( Tutorial (AI News) ). By using schema isolation, you can get just 2 databases to support 10-20 products! Plus, with 5-minute AI-assisted migration, you can max out your free tier. Pay just $25 and score dozens of times the quota – a total bang for your buck! 💰

4. OpenClaw suggests the “session” concept should disappear. OpenClaw might just be the death knell for the ‘session’ concept 🔥, according to one hot take ( Discussion (AI News) ). Constantly starting new conversations forces users to work around machine limitations, and the paradigm is shifting towards contextual fluid collapse. Agents can only achieve a continuous life trajectory by shaking off the legacy of sessions.

5. AI views literary style as merely a probability table. Spark Seed shared a fascinating thought 💭: humans summarize to fight forgetting, while AI summarizes to fight chaos ( Original Post (AI News) ). From an AI’s perspective, Lu Xun’s literary style is just a modified probability table and word gravity parameters. Literary soul, reduced to a digital fingerprint – harsh, but true.

AI News Daily Audio Version

🎙️ Xiaoyuzhou	📹 Douyin
Laisheng Xiaojiuguan	Self-Media Account