02-18-Daily AI News Daily

AI News Daily 2026/2/18

AI Info | Daily Read | Web Data Aggregation | Frontier Science Exploration | Industry Open Mic | Open-Source Innovation Power | AI and Human Future | Visit Web Version | Join Group Chat

Today’s Digest

Doubao AI interactions hit 1.9 billion on New Year's Eve, Seedance 2.0 launched; Grok 4.2 released
Frontier research emerges: multimodal "modal aphasia," geometric classification of LLM hallucinations
Galaxy Universal Robot debuts at Spring Festival Gala, powered by end-to-end full-body large model AstraBrain
Open-source projects like Superpowers and Hummingbot skyrocket in popularity
Discussions ignited on AI cognitive debt, regulatory disputes, and Spring Festival Gala's commercial value dwarfing AI companies' ARR

Product & Feature Updates

1. Doubao AI interactions hit 1.9 billion on New Year’s Eve. Doubao, ByteDance’s platform, teamed up with the CCTV Spring Festival Gala, rolling out tons of interactive features. On Chinese New Year’s Eve, it generated over 50 million New Year avatars. At its peak, it hit a mind-blowing inference throughput of 63.3 billion tokens per minute. Its latest video model, Seedance 2.0, provided visual creative support 💡 for several gala programs Details in Synced Review Report (AI Info) . Both Seedance 2.0 and Doubao 2.0 Pro are now fully live 🚀.

2. Grok 4.2 Officially Launched. Grok 4.2, xAI’s model, just got a fresh new update 💡 and is now available to users. While detailed benchmarks for its improved capabilities haven’t been released yet, a Source Tweet (AI Info) shows a launch screenshot. Its future performance evaluations are definitely something to keep an eye on 🔥.

3. Ant Afu Debuts on Spring Festival Gala Stage. Ant Afu, a free health management app, made an appearance in the sketch “Blood Pressure Monitor” at the Spring Festival Gala. It connects 5,000 hospitals nationwide and topped the App Store download charts for three consecutive days during the Spring Festival 💥. A new trend is emerging with young people returning home and teaching their families to manage health using Ant Afu 🔥 See Report (AI Info) . New users can snag a 16.8 yuan cash bonus.
   
   

Frontier Research

1. Multimodal Models Suffer from ‘Modal Aphasia’. Multimodal models have been found to precisely reproduce movie posters visually 🎨, but they mess up crucial details when asked to describe them in text. This systemic disconnect between visual memory and textual expression has been dubbed Modal Aphasia. Even more alarming, safety guardrails might be bypassed through cross-modal interactions ⚠️ Paper Link (AI Info) . Models aligned only on text can still generate unsafe images.

2. LLM Hallucinations Get Geometric Classification. Researchers have proposed a geometric definition for three types of LLM hallucinations 🧠: Unfaithful (ignoring context), Inventive (fabricating content), and Factually Incorrect (correct framework, wrong details). While detection of artificially invented content can hit 0.96 AUROC 💡, factually incorrect hallucinations are nearly indistinguishable in the embedding space Paper Details (AI Info) .

3. Smart Contract Translation Gets End-to-End Agent Pipeline. A research team has built an automated framework for translating natural language into Solidity code 🔗, creating an end-to-end agent pipeline for smart contract translation. The system uses a CrewAI-style team of agents for iterative refinement. Quality assessment spans five dimensions, including functional completeness and state machine correctness. This framework also supports comparative scoring against standard implementations 💡 Paper Address (AI Info) , providing a reproducible benchmark for smart contract synthesis research 🚀.

4. Virtual Try-On Achieves Training-Free Universal Framework. OmniVTON++ works without needing retraining on specific data 🎯. It incrementally adjusts human body structures during diffusion sampling via main pose guidance. Evaluations across datasets and clothing types show optimal performance ✨ Paper Details (AI Info) . It even supports multiple people, multiple garments, and anime character try-ons. The source code is set to be publicly released 🔥.

Industry Outlook & Social Impact

1. Galaxy Universal Robot ‘Xiaogai’ Goes Viral for Spring Festival Gala Work. Galaxy Universal Robot “Xiaogai” 🤖, the “iron buddy” in Shen Teng’s skit, went viral at the Spring Festival Gala, first cracking walnuts and then retrieving goods on stage. It’s powered by AstraBrain, the world’s first end-to-end full-body large model. Galaxy Universal has already rolled out over 100 unmanned retail stores nationwide 💡. The company, founded just two and a half years ago, has raised over $800 million and secured thousands of orders from giants like CATL and Toyota Details (AI Info) . We’re seeing a leap from “can demonstrate” to “can actually get work done.”

2. Andrew Ng Criticizes Excessive EU AI Regulation. Andrew Ng has slammed excessive EU AI regulation 🚫, stating that slowing down won’t lead the AI revolution. He believes the EU AI Act is stifling innovation 💀. These comments came during a DeepLearning.AI event speech. The AI Dev 26 conference will continue this discussion in late April in San Francisco 🌉 Speech Video (AI Info) . The balance between regulation and innovation remains a global conundrum ⚖️.
   
   

3. Long-Term Unemployment Becoming New Normal in Labor Market. Long-term unemployment is becoming the new normal in the labor market, with approximately 1.8 million Americans searching for jobs for over six months without success 😩. Commentators point out that official BLS data is often significantly revised downwards 📉. AI automation and offshore outsourcing are seen as structural drivers. H-1B holders and contract workers face greater risks 🔥 Discussion Details (AI Info) , and the accelerated loss of entry-level jobs is sparking widespread concern.

4. AI Executives’ Fear Marketing Sparks Community Division. AI executives’ fear marketing, including Microsoft execs and Sam Altman’s doomsday narratives 💀, is being criticized as FOMO-driven. Commentators believe these talking points are primarily aimed at CEOs and investors 🎯. Many enterprise-level AI projects have yet to show significant ROI. Developers admit LLMs speed things up but can’t replace deep engineering judgment ⚠️ Discussion Post (AI Info) . Geopolitical arms race anxieties are also frequently brought up.

5. Research Claims Self-Generated Agent Skills Ineffective, Sparks Controversy. Research claims that self-generated agent skills (where models write their own skill documentation) actually degrade performance 📉, sparking controversy. Critics, however, slam the experimental setup for prohibiting internet access and trial-and-error 🚫, calling it unreasonable. Practitioners commonly use an “execute → observe → refine” closed-loop approach. Manually curated skills, for example, boosted performance by +51.9pp in the medical field 🚀 Discussion Details (AI Info) . The disconnect between academic research and engineering practice is once again laid bare 💡.

6. WebMCP Proposal Allows Webpages to Expose Tool Interfaces to LLMs. The WebMCP proposal, which aims to let webpages expose tool interfaces to LLMs, has seen an early preview release from Chrome 🧪. However, the security and privacy section was left blank, drawing mockery 😅. Prompt injection and permission abuse risks are repeatedly highlighted ⚠️. The alternative, SKILL.md, is gaining support due to its simple deployment Proposal Discussion (AI Info) . The trade-off between accessibility and re-definition remains unresolved 💡.

7. Show HN Flooded with AI-Assembled Projects. Show HN is getting flooded with tons of “vibe coding” projects rapidly generated by LLMs 🌊. Screen recording tools and blog templates are major hotspots 💤. However, some shared cases of earning $6 million in revenue through Show HN. The community suggests creating a dedicated section to distinguish AI-generated products from thoroughly developed projects 🏷️ Discussion Details (AI Info) . Platform governance and signal dilution issues are gaining widespread attention.

Top Open-Source Projects

1. Superpowers: Terminal Enhancement Toolset. Superpowers, a Shell project with ⭐53.2k stars, has become GitHub’s hottest item today 🔥. It racked up 725 new stars today, providing a one-stop enhancement capability for command-line workflows 💡 Project Address (AI Info) . With 4045 forks, community engagement is clearly through the roof.

2. Hummingbot: Crypto Algorithmic Trading Bot. Hummingbot, a Python open-source project, has already amassed ⭐16.8k stars. Today, it surged with 565 new stars 🚀, showing fierce momentum. It supports automated market-making strategies across various exchanges 💰 Repository Link (AI Info) . A whopping 4411 forks indicate a super active quant community 😎.

3. Seerr: Media Request Management Platform. Seerr, a TypeScript project with ⭐9.2k stars, gained 283 new stars today. It offers intelligent request management for home media servers 🎬 Project Address (AI Info) . With 609 forks, it shows a steadily growing user base 📈.

4. Heretic: Emerging Python Open-Source Project. Heretic, a Python open-source project, saw an astonishing 666 new stars today 🔥. The project, with ⭐6.8k stars, is in a rapid growth phase 🚀 Repository Address (AI Info) . Its 684 forks suggest significant community interest 💡.

5. Claude-Quickstarts: Official Anthropic Quick Start. Claude-Quickstarts, a Python project with ⭐14.3k stars, is maintained by the Anthropic team 🏢. It offers various integration examples and best practices for Claude models 💡 Project Address (AI Info) . Gaining 102 new stars today and boasting 2424 forks, it’s super practical 🔧.

6. OpenCTI: Open-Source Cyber Threat Intelligence Platform. OpenCTI, a TypeScript project with ⭐8.3k stars, focuses on the security domain 🛡️. It provides structured threat intelligence management capabilities for organizations Repository Link (AI Info) . With 1222 forks, it clearly has a solid foundation in the security community 💡.

7. Shard: Distributed P2P AI Inference Network. Shard, a distributed P2P AI inference network, has been built by developers who crafted browser-side inference nodes based on WebGPU 🌐. It runs with a libp2p mesh network and a Rust daemon. The goal is to explore whether decentralized inference can reduce reliance on GPU cloud services 💡 Project Address (AI Info) . Currently in its early stages ⚠️, technical feedback is welcome!

Social Media Share

1. OpenClaw Deployment Barrier Lowered by Baidu. OpenClaw deployment just got way easier, thanks to Baidu. The Baidu App has integrated OpenClaw, allowing for installation with just a phrase in the search bar. The blogger suggests not stressing about servers or Mac mini for now 💡. Plus, the OpenClaw founder officially announced joining OpenAI today 🥳 Tweet Details (AI Info) . This might just be the simplest deployment method available right now 🚀.

2. Three.js Author Also Chose OpenClaw. mrdoob, the author of Three.js, publicly stated he’s using OpenClaw paired with MiniMax 2.5 🔥. This choice has grabbed the attention of the developer community 💡 Source Tweet (AI Info) . An endorsement from a top open-source author is super persuasive 👍. MiniMax 2.5’s performance is getting real-world validation 🚀.

3. Building Multi-Agent Systems with Codex is Like Play. Building multi-agent systems with OpenAI Codex 🤖 is apparently like play! A blogger shared their experience, with interface screenshots showing an incredibly simple process 💡 Tweet Details (AI Info) . The barrier to agentic coding is dropping rapidly 🚀, and the results are so smooth, it’s unreal ✨.

4. AI Helps Repay Technical Debt But Accelerates Cognitive Debt. AI, while helping to repay technical debt, might actually accelerate cognitive debt. A blogger cited Peter Naur’s view that “programming is theory” 💡, suggesting that even with clear AI-generated code, developers might completely lose their understanding of the system ⚠️. Team case studies show cognitive debt can totally cripple projects 😩 Blog Post (AI Info) . The recommendation? At least one human should fully grasp every AI-generated change 🧠.

5. Spring Festival Gala’s Commercial Value Crushes Chinese AI Companies’ ARR. The Spring Festival Gala’s commercial value is absolutely crushing the combined Annual Recurring Revenue (ARR) of Chinese AI companies. A Jike user pointed out that CCTV’s Gala raked in over 5 billion yuan in sponsorship fees 💰, with Doubao hitting a peak daily active user count of over 200 million 🥳. All disclosed ARR from Chinese AI companies combined can’t even touch that single day 🤯 Source (AI Info) . Robot concept stocks, meanwhile, are seeing a solid bump 📈.

6. Unitree Robotics’ Strategic Approach Affirmed. Unitree Robotics’ strategic approach has been affirmed by a blogger, who noted that the company focuses on perfecting the motion platform 🤖 rather than just chasing intelligence. They prioritize perfecting joints and processes to the extreme 💪. Making movements incredibly fluid is the right product and marketing strategy 🎯 Tweet Details (AI Info) . The evolution from the 2025 to 2026 Spring Festival Gala is strikingly evident ✨ Unitree Robotics Spring Festival Gala Performance Video (AI Info) .
   
   

7. Growth Path from 0 to 30,000 Followers on X. The growth path from 0 to 30,000 followers on X involves several stages. During the cold start, it means leaving 30-50 high-quality comments daily under large accounts 💬. Growth from 500 to 5,000 relies on gift lead magnets 📦. From 5,000 to 10,000, it’s about reverse-engineering viral content structures 🔍. The scaling phase requires high-frequency output of 5-10 posts daily 🚀 Tweet Details (AI Info) . The hardest part is getting from 0 to 500, where most people give up 😥.

8. The ‘Internet is Dead’ Trilogy Serialization Completed. Orange AI has wrapped up its “Internet is Dead” trilogy, a series set in an Agent world view 💡. The third installment is super human-heavy 🤗. The new series will draw inspiration from “Love, Death & Robots” to explore short stories across different dimensions 🎬 Series Summary (AI Info) . Happy Chinese New Year on the first day of the lunar calendar! 🧧.

9. 10 AI Cognitive Insights Attract Attention. Blogger Tong Ouba spent a whole year’s worth of tokens to uncover 10 AI cognitive insights that challenge intuition 🔥. These insights might be counter-intuitive for many 💡, but they’re precisely what create the difference ✨ Article Link (AI Info) . Here’s to everyone taming their AI steeds in 2026! 🐎.

10. Twitter Also Crashed on New Year’s Eve. Twitter also crashed on New Year’s Eve, leading a blogger to gripe about whether Elon Musk’s 30 tech staff could maintain it 24/7 😩. Some joked that the Spring Festival Gala was better than Twitter that night 🎭 Source Tweet (AI Info) . Platform stability issues were exposed yet again 🚨, and the cost of a lean team is glaringly obvious 💀.

AI News Daily Audio Version

🎙️ Xiaoyuzhou	📹 Douyin
Afterlife Tavern	Self-Media Account