Top Stories
India AI Summit Hosts First Major Global South Research Symposium
The Research Symposium on AI and Its Impact began today at Bharat Mandapam, serving as the principal academic platform of the India-AI Impact Summit 2026. IIIT Hyderabad partners with the summit on what’s being called the first academic AI forum of this scale in the Global South.
The symposium received around 250 research submissions from Africa, Asia, and Latin America. Sessions focus on AI-driven scientific discovery, safety and governance frameworks, equitable access to compute infrastructure, and collaborative research across developing nations. The format includes plenary sessions, international research presentations, and a Global South poster showcase.
The timing is deliberate: as Western AI governance fractures and China builds a separate ecosystem, the Global South is positioning itself as a third pole. Whether this produces meaningful coordination or just another set of competing frameworks depends on what comes out of these sessions.
Sources: India AI Summit, Government Press Release
Cohere Releases Tiny Aya: Open Multilingual Models That Run Offline
Cohere launched Tiny Aya, a family of open-weight multilingual models supporting 70+ languages and capable of running on everyday devices without internet. The base model contains 3.35 billion parameters - small enough to run locally, powerful enough for serious work.
The release includes regional variants: TinyAya-Earth for African languages, TinyAya-Fire for South Asia (Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, Marathi), and TinyAya-Water for Asia Pacific and Europe. All variants are available on HuggingFace, Kaggle, and Ollama.
Notably, the models were trained on a single cluster of 64 H100 GPUs. That’s modest compute for a multilingual model of this capability. Cohere announced the release at the India AI Summit, timing that underscores the model’s focus on Global South languages often underserved by larger labs.
For privacy-conscious users and developers in low-connectivity regions, this matters: capable local AI that doesn’t phone home.
Sources: TechCrunch, Dataconomy
Apple Announces March 4 “Special Experience” Across Three Continents
Apple is holding a “Special Apple Experience” on March 4 in New York, London, and Shanghai simultaneously. The word choice - “experience” rather than “event” - suggests something different from the usual live-streamed announcements from Apple Park.
Mark Gurman reports that product announcements will come via press releases across Monday, Tuesday, and Wednesday of that week rather than a single keynote. Expected hardware includes iPhone 17e, MacBook Pro with M5 Pro/Max chips, eighth-generation iPad Air, and possibly a new low-cost MacBook with an A18 chip.
The globally synchronized launch format is new for Apple. Mobile World Congress happens the same week, creating a dense news cycle. Whether this is Apple learning to compete for attention in an AI-saturated news environment or just experimentation remains unclear.
Sources: MacRumors Event Announcement, MacRumors Preview
Quick Hits
-
Gemini “mature enough to take over Search”: Bloomberg reports Google is rapidly transitioning from traditional Search to AI-enhanced Gemini. The chatbot now has 750M monthly active users, and Gemini 3 powers both AI Mode in Search and Google’s enterprise offering with 8 million paying licenses. Bloomberg
-
International AI Safety Report warns of insufficient safeguards: The second annual report from 100+ experts across 30 countries finds AI capabilities advancing faster than safety measures. ChatGPT now has ~700M weekly users (up from 200M a year ago), but “no combination of current methods eliminates failures entirely.” Leading models passed medical/law licensing exams but still fabricate information and produce flawed code. Full Report
-
DeepSeek V4 still pending: Despite yesterday’s speculation about a Lunar New Year release, DeepSeek V4 hasn’t dropped yet. The company maintains operational silence on specific dates. When it does arrive, expect 1M+ token context, the new Engram architecture, and claimed performance exceeding Claude 3.5 Sonnet on coding benchmarks - all on consumer hardware. Introl
-
OpenAI’s Codex-Spark hits 1000 tokens/second: The new real-time coding model, released last week on Cerebras hardware, is OpenAI’s first model not running on Nvidia. Available to ChatGPT Pro users through the Codex app and VS Code extension, it trades some capability for near-instant response. OpenAI
Worth Watching
The India AI Summit’s research symposium is the substantive counterweight to the main event’s political theater. If the 250 submitted papers and resulting collaborations produce actual infrastructure-sharing agreements or safety framework coordination, that’s more meaningful than another joint declaration. Watch for announcements on compute access - Cohere’s timing suggests they’re betting on this audience.
Google’s transition to AI-first Search is the bigger long-term story. With 750M monthly Gemini users and the Bloomberg assessment that Gemini is “mature enough to take over,” we’re watching the potential end of traditional search. The advertising implications alone could reshape how the entire web works.
Apple’s March 4 multi-city format is interesting tactically. They’re competing for attention in a week that includes MWC and whatever AI announcements get timed around it. The distributed press-release approach may signal how hardware companies adapt to an AI-dominated news cycle.