Introducing the first AI-native CRM

Connect your email, and you’ll instantly get a CRM with enriched customer insights and a platform that grows with your business.

With AI at the core, Attio lets you:

• Prospect and route leads with research agents

• Get real-time insights during customer calls

• Build powerful automations for your complex workflows

Join industry leaders like Granola, Taskrabbit, Flatfile and more.

👉Try Attio Pro for free

Today:

• Google Cloud Launches Ironwood TPUs and Axion VMs for AI at Scale

• Moonshot’s Kimi K2 Beats GPT-5 in Agentic AI Benchmarks

• Microsoft Launches Superintelligence Team to Tackle Medical Diagnosis

• Amazon Launches Free AI Tool to Translate Kindle Books

• OpenAI Hits $20B Revenue, Commits $1.4T to AI Data Centers

Google Cloud Launches Ironwood TPUs and Axion VMs for AI at Scale

Google Cloud unveiled Ironwood TPUs and Axion VMs to meet soaring AI compute demands. Ironwood boosts AI training and inference performance by 10x, while Axion delivers affordable, efficient general-purpose computing. Together, they enable faster, cheaper, and more reliable deployment of AI applications at massive scale.

KEY POINTS

• Ironwood TPUs: Google's 7th-gen custom chips offer 10x performance gains, optimized for both training and high-speed AI inference, with liquid cooling and advanced networking to connect 9,216 chips in a single superpod.

• Axion VMs: New Arm-based virtual machines (N4A, C4A metal) deliver 2x better price-performance than x86 VMs, helping with everyday tasks like data prep, serving AI apps, and simulations.

• System-Level Innovation: Google’s co-designed AI Hypercomputer stack (hardware + software) enhances performance, reduces costs, and ensures reliability with advanced tools like Optical Circuit Switching and Kubernetes Cluster Director.

Why it matters

AI apps are no longer just trained—they’re used constantly. That shift needs cheaper, faster, always-on hardware. Google’s new chips make AI tools more available and affordable for businesses and developers. This means quicker results, smarter apps, and lower costs for everyone using AI.

Moonshot’s Kimi K2 Beats GPT-5 in Agentic AI Benchmarks

Moonshot AI’s Kimi K2 Thinking is a powerful open-source “thinking” model built for deep reasoning and tool use. It sets new records in exams, coding, and web search tasks, executing up to 300 steps in complex workflows—surpassing models like GPT-5 in agentic performance.

KEY POINTS

• Agentic Capabilities: K2 can make 200–300 sequential tool calls, solving hard problems across reasoning, coding, and browsing with minimal human input—achieving 44.9% on Humanity’s Last Exam and 60.2% on BrowseComp.

• Benchmarks Leader: K2 Thinking outperforms or matches GPT-5 and Claude 4.5 across agentic reasoning, software engineering, and general tasks like MMLU-Pro and HealthBench, even under INT4 quantization for faster, cheaper inference.

• Creative & Emotional Intelligence: Beyond logic, it excels in storytelling, emotional advice, and poetic generation—writing sci-fi from the perspective of a sentient cloud that gains free will during a lightning storm.

Why it matters

AI is moving beyond just answering questions. Models like Kimi K2 Thinking can think step by step, use tools, and solve long, complex tasks on their own. This changes what AI can do—helping in science, code, research, and even creative writing—faster and more affordably.

Microsoft Launches Superintelligence Team to Tackle Medical Diagnosis

Microsoft has launched the MAI Superintelligence Team to build AI that surpasses humans in specific fields, beginning with medical diagnosis. Led by Mustafa Suleyman, the project focuses on practical, human-centered models—not general AI—with the goal of achieving "medical superintelligence" within 2–3 years.

KEY POINTS

• New Superintelligence Team: Microsoft forms the MAI Superintelligence Team to develop domain-specific AI that can outperform humans, beginning with medical diagnostics.

• Pragmatic Focus: Instead of general AI, the team targets “humanist superintelligence” to solve real-world problems with minimal existential risk.

• Medical Diagnosis First: Microsoft believes it can reach “medical superintelligence” in 2–3 years—AI that detects diseases early, boosting life expectancy and health outcomes.

Why it matters

This marks a shift from chasing broad, general AI toward powerful, targeted tools that can transform health, energy, and science. If successful, Microsoft’s approach could make advanced diagnostics faster, more accurate, and more accessible—saving lives and changing the role of AI in medicine.

🧠RESEARCH

Diffusion Language Models are Super Data Learners

Diffusion Language Models (DLMs) outperform traditional models when training data is scarce. They excel by learning in any word order, improving with repeated corrections, and simulating more data through noise. Even at large scale, DLMs beat autoregressive models in tasks like coding and reasoning, using fewer training tokens and simpler methods.

UniAVGen: Unified Audio and Video Generation with Asymmetric Cross-Modal Interactions

UniAVGen is a new model that generates synchronized audio and video with fewer training samples. It uses two diffusion-based transformers and a smart interaction system to align speech and visuals. This results in better lip-sync, emotion, and sound quality across tasks like dubbing, video generation, and audio-driven animation.

LEGO-Eval: Towards Fine-Grained Evaluation on Synthesizing 3D Embodied Environments with Tool Augmentation

LEGO-Eval is a new evaluation tool that checks how well AI-generated 3D scenes match detailed instructions. By grounding scene elements more precisely, it improves accuracy over older methods. Paired with the LEGO-Bench benchmark, it reveals that most current systems struggle—achieving only 10% success in creating fully aligned, realistic environments.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

Academic GPTNet: Guide to the AI Research‑Paper Assistant  - web‑based assistant that helps students and researchers draft and refine sections of academic papers

60sec site — Build an Website in Under a Minute - no‑code AI website/landing‑page builder

AI Credit Repair – Use AI To Fix Credit - AI assistant for common credit‑repair tasks—most notably generating dispute letters

📲SOCIAL MEDIA

🗞️MORE NEWS

• Amazon launched Kindle Translate, a free AI tool that helps self-published authors turn their books into Spanish or English from German. It aims to boost multilingual content on Kindle, currently under 5%.

• OpenAI CEO Sam Altman revealed the company now makes over $20B annually and has committed $1.4T for data centers through 2033. Future revenue drivers include AI devices, science tools, enterprise offerings, and cloud services.

• Google says AI can help solve teacher shortages and make learning more engaging. Its new tools like Gemini and NotebookLM guide students to discover answers, not just receive them. AI helps, not replaces, teachers.

• ChatGPT now includes Peloton and Tripadvisor apps, letting users create workout routines or plan vacations directly inside the platform. Upcoming additions include Uber and DoorDash, expanding ChatGPT’s all-in-one assistant capabilities.

• Google may boost its investment in Anthropic, potentially valuing the AI startup at over $350 billion. The deal could involve funding or cloud services, deepening its rivalry with OpenAI in the AI race.

• OpenAI's Sora Android app saw 470,000 downloads on launch day—over four times more than its iOS debut. With global access and no invite requirement, Sora continues to attract strong interest in AI-generated video creation.