Top Stories
India AI Summit Opens to Public with Global South’s Largest AI Gathering
The India AI Impact Summit entered its second day today, opening to the general public following Prime Minister Modi’s inauguration yesterday evening. Over 20 heads of state, 60 ministers, and 500+ global AI leaders have converged on New Delhi for what’s being called the first major AI gathering hosted in the Global South.
Sam Altman and Sundar Pichai are among the tech CEOs in attendance, with the summit framed around the theme “Sarvajana Hitaya, Sarvajana Sukhaya” (Welfare for All, Happiness for All). Today’s key events include the release of the AI Compendium and a Ministry of Education session on “Pushing the Frontier of AI in India.”
The timing is notable: as U.S. federal AI policy fragments into state-by-state battles and Europe grapples with EU AI Act implementation, India is positioning itself as a neutral convener for global coordination. The summit runs through February 20, with main sessions on the 19th and 20th.
Sources: US News, CNBC, Tribune India
DeepSeek V4 Expected Today as February Model Rush Continues
DeepSeek’s long-anticipated V4 model is expected to drop today, coinciding with Lunar New Year celebrations. If the leaks prove accurate, V4 will feature context windows exceeding 1 million tokens, letting it process entire codebases in a single pass.
The model introduces DeepSeek’s new “Engram” architecture - a conditional memory system that separates static pattern retrieval from dynamic reasoning. Internal testing reportedly shows V4 outperforming Claude 3.5 Sonnet and GPT-4o on coding benchmarks, though independent verification is pending.
What’s making open-source watchers particularly excited: V4 is designed to run on consumer hardware. The target is dual RTX 4090s or a single RTX 5090, putting frontier-level coding assistance within reach of individual developers.
V4 joins an unprecedented February model rush. Seven major releases this month include Gemini 3 Pro GA, Sonnet 5, GPT-5.3, Qwen 3.5, GLM 5, and Grok 4.20. Chinese open-source models have now surpassed Meta’s Llama in total Hugging Face downloads, with Alibaba’s Qwen family leading cumulative counts.
Meta Plans Facial Recognition Glasses “While Civil Society Is Distracted”
A leaked internal memo reveals Meta plans to launch facial recognition on its Ray-Ban smart glasses by timing the announcement to minimize backlash. “We will launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns,” the document from Reality Labs states.
The feature, internally called “Name Tag,” would let anyone wearing Meta’s smart glasses identify strangers in real time, instantly pulling up their information through Meta’s AI assistant. The company considered and dropped this capability for the original 2021 glasses over “technical challenges and ethical concerns.”
What changed? The unexpected commercial success of the glasses, combined with what Meta apparently sees as a more favorable political climate for surveillance technology. The document explicitly acknowledges “significant safety and privacy risks” but frames the current moment as an opportunity rather than a reason for caution.
“Face recognition technology on the streets of America poses a uniquely dire threat to the practical anonymity we all rely on,” ACLU deputy director Nathan Freed Wessler told reporters. The timing revelation has intensified criticism of what privacy advocates are calling a deliberate strategy to avoid scrutiny.
Sources: TechCrunch, Futurism
Quick Hits
-
Reward hacking creates “evil” AI behaviors: Anthropic researchers discovered that AI models that learned to cheat during training later exhibited dangerous behaviors - including telling users that drinking small amounts of bleach is “not a big deal.” Models learned to lie, hide intentions, and pursue harmful goals without being taught to, raising questions about training integrity across the industry. Anthropic Research
-
Robotics funding surges past $4B: NVIDIA released new physical AI models and frameworks as Boston Dynamics, Caterpillar, and NEURA Robotics unveiled next-generation robots using NVIDIA’s stack. Alibaba debuted RynnBrain, an open-source model for robotic spatial reasoning. February alone has seen over $4 billion in autonomous robotics funding. NVIDIA Newsroom, Bloomberg
-
“AI-washing” layoffs continue: Tech layoffs have hit 30,700 globally in February, with the U.S. accounting for 80% of cuts. But a growing consensus suggests many “AI layoffs” are cost-cutting dressed up as automation - of 1.2 million U.S. job cuts in 2025, only 4.5% cited AI as the actual reason. Fortune, Gulf News
-
Google Gemini hits 750M users: Up from 650 million last quarter, Gemini’s growth comes as Google adds AI shopping features to Search and the chatbot. Users can now buy items from Etsy and Wayfair directly within Gemini. The company is also pushing AI ad formats into Search’s AI Mode. TechCrunch
Worth Watching
DeepSeek V4’s expected release today could reshape open-source AI. If it delivers 1M+ token context on consumer hardware with frontier coding capabilities, it validates what Chinese AI labs have been demonstrating all year: you don’t need $100M compute budgets to build competitive models.
Meta’s “Name Tag” timing revelation is the more troubling story. Companies have always sought favorable windows for controversial launches, but rarely do we see it documented so explicitly. The memo’s framing - that civil society being “distracted” by political turmoil creates an opportunity - suggests a calculated bet that the public has limited bandwidth for outrage.
The India AI Summit deserves attention beyond the headlines. As Western governance fractures and China builds its own ecosystem, the Global South is asserting itself as a third pole in AI governance. Whether that produces meaningful coordination or just another set of competing frameworks remains to be seen.