Punjab government approves a government GPU cloud project to build a local AI platform for digital governance. The system will enable hosting and training large language models locally, support Urdu and other languages, and provide secure sandboxes for institutions and universities. Officials say it improves data security and boosts local tech capacity. #Punjab #AI #CloudComputing #DigitalPakistan #TechNews

🚀 Subquadratic outperforms FlashAttention-2 in speed 📈 Appen’s report reveals that its Sparse Self-Attention kernel achieves 381 ms on 1 million tokens. FlashAttention-2 takes 21.4 seconds. 62.8× reduction in FLOPs. Maintains high performance in quality and retrieval. A revolution in AI efficiency is possible.

Disguising cuts as upgrades, calling AI price hikes “subsidies”... 😡 Claude’s parent company, @AnthropicAI, just announced that starting June 15, paid Claude plans will have a separate, programmatic usage allowance (for claude-p, Agent SDK, GitHub Actions, etc.). Sounds like a gift? It’s effectively a price increase disguised as free credits 💸 Previously, these tools shared the same usage quota as chat, with a 5x–20x subsidy. Now, with them separated, Pro gets only $20, and Max’s 20x multiplier is capped at $200—once exhausted, you pay full API rates. For existing developers, this means costs have skyrocketed 5–10x, sparking outrage across the community. Developers are calling it outrageous to label this downgrade as “free credits.” Many are declaring they’ll switch to OpenAI Codex or simply discontinue third-party tools. The only positive feedback comes from light users who say they “got a little bonus”—but these users were never heavy Anthropic developers anyway. Email notifications will go out on June 8; the changes take effect on June 15.

Codex has recently rolled out a series of updates—GPT-5.5, Plan Mode, and Goals for long tasks—narrowing the gap with Claude rapidly, and even surpassing it in some scenarios. The agent operation experience is also becoming increasingly similar, making Xiao Long Xia consider stepping back. In particular, the combination of Goals and Computer Use greatly enhances the validation of complex frontend systems, creating a more cohesive workflow that links requirements, execution, and verification. Claude has already downgraded to Pro; we’ll wait to see how the much-hyped Mythos performs before making further decisions.

WeChat chat history now supports one-tap forwarding to Yuanbao, enabling AI to instantly generate content summaries and significantly improve work efficiency. WeChat has officially integrated cross-app message processing with Tencent’s native AI assistant, Yuanbao. Now, when faced with group chats containing 99+ unread messages or lengthy private conversations, users can simply select the chat logs they wish to summarize, tap “Forward to Another App,” and send them to Yuanbao—allowing the AI to generate a concise summary with one click. In addition to its core “summary extraction” feature, this update includes several practical use cases built around extracted chat data. For example, forwarding invoices or expense records to Yuanbao will automatically generate a table listing total amounts and purposes; sending lengthy group discussions about travel plans allows Yuanbao to identify key requirements and create personalized itineraries; in private chats, it can even analyze conversation dynamics and suggest appropriate responses for users engaging with someone they’re interested in. This design leverages common pain points in WeChat usage to provide a seamless and natural entry point for Tencent’s standalone AI app, Yuanbao. Currently, users must update both the WeChat and Yuanbao apps to their latest versions to access this feature. #AI #AIAgent

Breaking🚨 Apple has the MacBook. Microsoft pushed Copilot+ PCs. Now Google is entering the race with something entirely different: Googlebook. But unlike everyone else, Google didn’t just add AI into a laptop. They built the whole experience around it 👇 → Magic Pointer Shake your cursor anywhere on screen and Gemini instantly pops up to help with whatever you’re looking at. → AI-built dashboards Tell Gemini what you need and it creates custom widgets for flights, reminders, schedules, and more in one workspace. → Deep Android integration Your phone and laptop finally work like one system. Apps, files, notifications — all synced automatically. → Built with major hardware brands Acer, ASUS, Dell, HP, and Lenovo are all onboard, each featuring Google’s new signature glowbar design. Chromebook was built for the cloud generation. Googlebook looks like Google’s answer to the AI generation. Expected later this year.

Gemini is handing out real phone numbers to strangers. DeleteMe says AI-related privacy queries up 400% in 7 months. Google took 47 days to reply to one victim. No opt-out exists. PII in training data is the exploit and nobody is patching it.

Anthropic will be giving Claude credits for third-party applications to all subscribers From June 15, subscribers will receive credits equal to the subscription amount, i.e., $20, $100, or $200, depending on the subscription tier. These credits can be used for applications based on the Agent SDK, such as OpenClaw, or for custom tools using Claude. This is what the life-giving SpaceX computer does

In Claude Code, a mode of managing multiple agents at once has been introduced, the number of open terminals will decrease significantly, and the developers of orchestrators on top of Claude Code have put in a lot of effort. In addition, the team from Anthropic has completely copied the mode from Codex /goal - in it, the model does not stop until it reaches its goal. I don't know who in Anthropic makes these announcement videos, but they're incredibly good