PostCards
On India's 80th Independence Day, Modi Commits to Training 1 Crore Youth in AI
Addressing the nation on its 80th Independence Day from the Red Fort in New Delhi, Prime Minister Narendra Modi announced a major push to train one crore — ten million — young Indians in artificial intelligence over the coming year, framing the country's youth as the driving force behind its Viksit Bharat ambitions.
Modi described the present moment as an era defined by robotics, AI, quantum technology, space and data centres, arguing that emerging technologies bring both new opportunities and new challenges that India's workforce must be equipped to meet head-on. The AI-skilling push was unveiled alongside a separate commitment to provide free online coaching for students preparing for competitive exams, aimed at easing the financial burden private coaching places on poor and middle-class families.
The announcements build on the momentum from India's hosting of the AI Impact Summit earlier this year and follow a broader "Sapt Dhara" roadmap for Viksit Bharat that the Prime Minister also outlined in the same address, spanning semiconductors, skilling and emerging-tech sectors.
IBM Strikes Strategic Partnership With OpenAI to Push Frontier Models Into the Enterprise
IBM announced a strategic partnership with OpenAI to help enterprises deploy AI at scale across core operations, embedding models including GPT-5.6, Codex and ChatGPT Work directly into IBM Consulting Advantage, the company's client-facing delivery platform.
The deal establishes a dedicated OpenAI practice inside IBM Consulting, with plans to train and certify tens of thousands of consultants over the coming months, primarily by retraining existing staff. The two companies will jointly build industry-specific solutions for financial services, government, telecommunications and retail, with particular emphasis on finance, procurement, customer operations and HR workflows.
The partnership also folds in the OpenAI Daybreak Cyber Partner Program launched earlier this year, integrating OpenAI's models with IBM's own multi-agent cybersecurity service. It arrives less than a year after IBM struck a similar alliance with Anthropic, and follows OpenAI's earlier consulting tie-ups with Infosys and Tata Consultancy Services — part of a broader push by OpenAI to reach enterprise customers through systems integrators rather than direct sales alone.
Apple Trains Its Own China-Only AI Model With Alibaba's Help — a First for a Foreign Firm
Apple has trained its own large language model specifically for the China market, developed with technical support from Alibaba, according to Reuters — a departure from the company's earlier strategy of leaning entirely on domestic partner models to power Apple Intelligence features there.
The move follows last month's approval from China's Cyberspace Administration, which cleared Apple's generative AI service for launch on Chinese iPhones. Alibaba's Qwen model will still be woven into Apple Intelligence across iOS, iPadOS, macOS and visionOS for Chinese users, while Apple separately works with Baidu on complementary features — described as a dual-track approach that lets Apple navigate regulatory hurdles that have blocked other US tech firms.
The rollout, expected in the coming months alongside a wider iOS update, would make Apple the first foreign company Beijing has approved to offer a proprietary AI model inside the country — a notable milestone given the deepening trade and diplomatic friction between Washington and Beijing over artificial intelligence, and as Apple works to fend off aggressive AI-phone competition from Huawei.
Anthropic Details How Claude's New Invisible Watermark Actually Works
Days after confirming that all Claude models launched on or after August 2 embed an imperceptible watermark into generated text, Anthropic published a fuller explainer detailing how the mechanism works, disclosing that it builds on an approach related to Google DeepMind's SynthID-Text and citing internal quality testing, including a live-traffic evaluation and a controlled human-rater study.
Generated text carries the watermark by subtly biasing word-choice patterns in a way that survives copy-and-paste, while supported file types such as SVG, PNG and JPG receive signed C2PA provenance metadata instead. Anthropic has confirmed a self-serve text-detection tool is coming, and that the underlying model itself has no awareness it is being watermarked.
The policy is Anthropic's response to Article 50 of the EU AI Act's Transparency Code, which took effect August 2 and carries fines of up to €15 million or 3% of global turnover for non-compliance — but the company has been explicit that the marking is applied worldwide, not just to EU traffic, and that neither a detected mark nor its absence is proof of who actually authored a piece of content.
OpenAI Ships "Ultrafast" Mode, Running GPT-5.6 Sol at 14x Normal Speed
OpenAI has rolled out a new processing mode called Ultrafast for GPT-5.6 Sol, its most capable model, claiming it can operate at up to 14 times the speed of standard processing — delivering as many as 750 output tokens per second.
The launch lands in the same week OpenAI shipped GPT-5.6-Cyber, a version of the flagship model with fewer restrictions on cyber-related content, and amid a broader industry shift where OpenAI and Anthropic are cutting prices even as rivals like DeepSeek move in the opposite direction, betting that faster, cheaper inference — not just smarter models — is what decides enterprise adoption.
Analysts tracking the shift note that as agentic workloads multiply, orchestration efficiency and reasoning-time economics may end up mattering as much to enterprise AI unit economics as the underlying model's raw benchmark scores.