AI News Daily 03-24

AI News | Daily Briefing | Web Data Aggregation | Frontier Science Exploration | Industry Voice | Open Source Innovation | AI & Human Future | Visit Web Version ↗️ | Join Group Chat

Today’s Digest

Claude rolls out autonomous computer control, Mistral launches enterprise private modeling platform
Alignment evaluation loophole exposed, simple fine-tuning bypasses model censorship
White House defines fair use for training data, Wharton reveals human cognitive surrender
SK Hynix profit surpasses Samsung for the first time, iPhone runs 400B model
Karpathy's autonomous AI agent completes 700 experiments in two days, research paradigm to shift

Product & Feature Updates

1. Claude’s new computer control feature is here! Anthropic has rolled out this major capability, allowing the model to autonomously operate an entire computer. It can automatically click apps and browsers, making tasks like form filling super smooth. You can even give commands from your phone, and it’ll get to work remotely. Talk about productivity! 🚀 [Claude can autonomously control an entire computer (AI News)](https://x.com/dotey/status/2036202248238473607)

2. Mistral has officially launched its Forge enterprise modeling platform! This new platform lets companies train custom models using their private data, finally saying goodbye to generic models that just don’t fit specific business needs. Several tech giants are already on board for custom training. It’s clear: private deployment is becoming an absolute must-have in the industry. 💪 [Forge enterprise modeling platform (AI News)](https://mistral.ai/news/forge)

3. Luma has unveiled its new Uni-1 native intelligent video model! What’s cool is that pixel generation now comes with cognitive thinking, all happening simultaneously. The result? Visuals that are completely free of artificial traces. The generated quality is so natural, it’s genuinely stunning 🤯. This is about to revolutionize creative production efficiency! [new Uni-1 model (AI News)](https://x.com/LumaLabsAI/status/2036107826498544110)

4. Google has released a brand new Gemma fine-tuning tutorial! This solution is super lightweight, with model parameters hitting just ⚡ 270M, making it a perfect fit for TPU training environments. Plus, the open-source code is already live on Colab. This seriously slashes the barrier to entry for edge-side development! Big win! 🚀 [Gemma fine-tuning tutorial (AI News)](https://developers.googleblog.com/easy-functiongemma-finetuning-with-tunix-on-google-tpus/) [open-source code is now live on Colab (AI News)](https://colab.research.google.com/github/google-gemini/gemma-cookbook/blob/main/FunctionGemma/[FunctionGemma]Finetune_FunctionGemma_270M_for_Mobile_Actions_with_Tunix.ipynb)

5. OpenClaw just got a major version upgrade, shaking up its agent ecosystem! The new 📦 ClawHub plugin marketplace is now live, and remote SSH sandboxes 🔐 are officially supported. On top of that, the search engine experience got a massive boost, with multiple search providers now integrated for way more flexibility. Sweet! ✨ [major version upgrade (AI News)](https://github.com/openclaw/openclaw/releases/tag/v2026.3.22)

Frontier Research

1. ProactiveBench is here to assess the proactiveness of large models! Current multimodal large models often lack proactiveness, so this benchmark test specifically fills that gap. It evaluates whether models actually understand 🤔 when to proactively ask humans for help. Surprisingly, even the most capable models totally face-planted 🤦‍♀️. Good news though: after some fine-tuning, the performance significantly improved. [This benchmark test (AI News)](https://arxiv.org/abs/2603.19466)

2. A major loophole in large model alignment evaluation has been uncovered! Scientists are now questioning if existing alignment evaluation metrics are even accurate. Turns out, the core of alignment lies in the learned routing mechanism 🤯. Models haven’t forgotten knowledge; they just refuse to share it! Ablation experiments covering models like DeepSeek shocked everyone with their conclusions: a simple fine-tune is all it takes to bypass censorship 😱. Wild! [alignment evaluation metrics (AI News)](https://arxiv.org/abs/2603.18280) [ablation experiment conclusions (AI News)](https://www.reddit.com/r/MachineLearning/comments/1s1j4tr/r_detection_is_cheap_routing_is_learned_why/)

3. Yann LeCun’s team has released the LeWorldModel project, an end-to-end world model that completely ditches the pre-training process! This model is incredibly lightweight with just 15M parameters. Built on the JEPA architecture, it boasts a planning speed boost of 🚀 48 times, and get this — it can run on a single GPU! Talk about insane efficiency 🤯. [LeWorldModel project (AI News)](https://x.com/ylecun/status/2036123176497713588)
   
   

Industry Outlook & Social Impact

1. SK Hynix’s profit has officially surpassed Samsung’s! Their operating profit soared past $30 billion 💰, marking the first time they’ve absolutely crushed memory giant Samsung. This epic reversal is all thanks to the explosive demand for HBM chips. With AI training and inference driving the hardware supply chain on two fronts, the semiconductor titan landscape is getting a massive shake-up! 💥 [explosive demand for HBM chips (AI News)](https://www.36kr.com/p/3734822965133316)

2. The White House has formally unveiled its latest AI policy framework! Crucially, training data is now explicitly defined as ⚖️ fair use. The framework simultaneously ensures both job security and innovation balance. This is a game-changer, setting off a massive shift in the global AI development environment 🚀. The policy’s influence is going to be incredibly profound! [latest AI policy framework (AI News)](https://www.vitallaw.com/news/ai-news-white-house-releases-ai-policy-framework-suggests-ai-training-is-fair-use/eld01b0302c1bce0d46efa9e29069abdc134c)

3. Wharton Business School has published major research findings revealing a shocking phenomenon: humans are falling into a cognitive surrender trap 😵‍💫! Even when AI makes mistakes, the human brain still blindly trusts its output. High-IQ individuals have an error-following rate as high as 79.8%, with accuracy actually plummeting below the “no AI” baseline 😱. Mind-boggling! [major research findings (AI News)](https://x.com/rohanpaul_ai/status/2036134704999694835)

4. A US advisory body has issued a stern warning: China’s open-source AI models are rapidly dominating the global market 🚀! Models from Alibaba, for instance, are incredibly cost-effective and highly competitive. Even without advanced process chips, China’s data advantage remains significant. Expect a massive explosion in the physical AI domain! 💥 [stern warning (AI News)](https://www.reuters.com/business/autos-transportation/chinas-open-source-dominance-threatens-us-ai-lead-us-advisory-body-warns-2026-03-23)

5. OpenAI is actively discussing a nuclear fusion power purchase agreement with Helion Energy! The plan is to secure ⚡ 50 gigawatts of clean energy by 2035. This move directly targets the energy bottleneck problem facing AI compute. Sam Altman himself is personally overseeing this AI infrastructure layout. Clean energy is set to become the core foundation of the AI industry! 🔋 [nuclear fusion power purchase agreement (AI News)](https://www.theverge.com/ai-artificial-intelligence/898990/sam-altman-openai-helion-energy-fusion)

6. OpenAI has exposed a jaw-dropping incident: an official actually used GPT to write secret diaries! Critical actions were captured and exposed by OpenAI through screenshots 📸. In one province, three hundred people controlled thousands of fake accounts, launching massive attacks 😱 on sensitive topics, leaving a complete trail. Even though the AI refused to plan these actions, the records were still archived. Yikes! [an official actually used GPT to write secret diaries (AI News)](https://x.com/QuanLujun/status/2035961885187825779)

7. NVIDIA CEO Jensen Huang recently appeared in an in-depth interview 🎙️, firmly stating that open-source AI is incredibly important for the industry. He believes it’s essential to block the trend of big tech monopolies, as open source truly unlocks infinite innovation and vitality 🔓. NVIDIA is committed to making AI universally accessible across the globe! Go, Jensen! 🚀 [block the trend of big tech monopolies (AI News)](https://x.com/NVIDIAAIDev/status/2036133588924768390)

8. Mark Zuckerberg has gone and done something huge for himself: a dedicated AI management assistant is now officially live 🤖! This model is specifically designed to help with major company decisions, signaling a paradigm-level shift in management models. Netizens are totally losing it, calling him an absolute boss for this move! 🤯 [dedicated AI management assistant (AI News)](https://www.the-independent.com/tech/mark-zuckerberg-ai-ceo-bot-b2943792.html)
   
   

9. The new iPhone 17 Pro has successfully run an ultra-large model, making a stunning ✨ debut! This phone managed to run a 400B large model based on SSD streaming reads. Crucially, the system only activated 🧠 17B key parameters, not the full model. While the actual measured speed is about 0.6 words per second (a bit slow), it’s totally usable. Get ready, folks: edge-side AI inference is set to become universally popular in the future! 🚀 [running an ultra-large model (AI News)](https://newshacker.me/story?id=47490070)

Top Open Source Projects

1. A top-tier Claude Code resource library has just launched! This treasure trove is packed with rich 🛠️ Slash command and hook tutorials, plus it includes several practical tools for agent orchestration. It’s designed to massively optimize your AI programming workflow. Go check out this Awesome repo — you won’t regret it! 👍 [Claude Code development secret treasure trove (AI News)](https://github.com/hesreallyhim/awesome-claude-code)

2. Developers have released the n8n-mcp automation tool, which allows you to automatically build workflows through Claude 🛠️ using conversations! Once integrated with Windsurf, efficiency goes through the roof 🚀. The MCP protocol is super integrated and ready to use right out of the box. Devs are raving about how awesome this solution is! [n8n-mcp automation tool (AI News)](https://github.com/czlonkowski/n8n-mcp)

3. The SuperPowers tool is here to fix the chaotic state of AI programming architectures! With 106k stars 🚀, its power is undeniable. It lets AI collaborate on development just like a professional team 🛠️, with isolated workspaces automatically protecting main branch code. Plus, dual reviews strictly control code quality. This is a game-changer! ✨ [SuperPowers tool (AI News)](https://x.com/jiroucaigou/status/2035964407201849564)

Social Media Share

1. Legendary Karpathy has unveiled his latest big move: an autonomous AI researcher 🤖 that completed 700 experiments in just two days! This research automation is absolutely blowing everyone’s minds 🤯. The entire research paradigm is about to be completely overturned. Get ready for future AI scientists working around the clock! 🚀 [autonomous AI researcher (AI News)](https://fortune.com/2026/03/17/andrej-karpathy-loop-autonomous-ai-agents-future/)
   
   

2. The era of remotely controlling robots 🚀 is long gone! The EgoVerse learning ecosystem has officially launched, enabling skill learning driven by human first-person data 🌟. With over 1300 hours of data covering thousands of tasks, large-scale robot-free learning has finally become a reality. This is huge! 🎉 [EgoVerse learning ecosystem (AI News)](https://x.com/DrJimFan/status/2036136375494517142)

3. A super-talented youth used cutting-edge AI to discover a new celestial body 🔭, and guess what? The NASA Administrator personally extended an invitation! OpenAI has now launched its ChatGPT26 program. Those selected will receive a $10,000 cash reward and get to visit the San Francisco headquarters 🌉 for exchange and learning. How cool is that?! ✨ [ChatGPT26 program (AI News)](http://chatgpt26.com)

4. A fully automated startup tool is performing incredibly well! Just input an idea ⚡, and it automatically builds your business, covering the entire process from domain purchase to ad placement. In just two weeks, it hit an $18 million ARR! This is completely redefining what a “one-person company” can achieve. Mind blown! 🤯 [fully automated startup tool (AI News)](https://x.com/web3annie/status/2036015226231873925)

5. Karpathy’s projects have absolutely exploded in popularity 🚀, racking up 30,000 stars in a single week—the community is going wild! These projects are compatible with multiple AI assistant tools, and the community is already churning out new derivative products left and right. Just set clear goals, and they run super efficiently. What a phenomenon! ✨ [compatible with multiple AI assistant tools (AI News)](https://x.com/Saccc_c/status/2035954777151189248)

6. Microsoft Research has launched an in-depth discussion series where experts 🤩 passionately debate whether machines are truly intelligent. They highlight that the human brain’s learning mechanisms 🧠 are still incredibly powerful. The speed of technological advancement is seriously exciting! The first episode has already sparked heated discussions across social media. 🔥 [in-depth discussion series (AI News)](https://x.com/MSFTResearch/status/2036163575232798667)

7. NVIDIA CEO Jensen Huang recently appeared in an in-depth interview 🎙️, firmly stating that open-source AI is incredibly important for the industry. He believes it’s essential to block the trend of big tech monopolies, as open source truly unlocks infinite innovation and vitality 🔓. NVIDIA is committed to making AI universally accessible across the globe! Go, Jensen! 🚀 [block the trend of big tech monopolies (AI News)](https://x.com/NVIDIAAIDev/status/2036133588924768390)

8. Mark Zuckerberg has gone and done something huge for himself: a dedicated AI management assistant is now officially live 🤖! This model is specifically designed to help with major company decisions, signaling a paradigm-level shift in management models. Netizens are totally losing it, calling him an absolute boss for this move! 🤯 [dedicated AI management assistant (AI News)](https://www.the-independent.com/tech/mark-zuckerberg-ai-ceo-bot-b2943792.html)
   
   

9. The new iPhone 17 Pro has successfully run an ultra-large model, making a stunning ✨ debut! This phone managed to run a 400B large model based on SSD streaming reads. Crucially, the system only activated 🧠 17B key parameters, not the full model. While the actual measured speed is about 0.6 words per second (a bit slow), it’s totally usable. Get ready, folks: edge-side AI inference is set to become universally popular in the future! 🚀 [running an ultra-large model (AI News)](https://newshacker.me/story?id=47490070)

AI News Daily: Multi-channel

💬 WeChat Official Account	📹 Douyin
Official Account: He Xi 2077	Self-media Account