Today:

• Elon Musk Unveils Grok 4, Teases ‘Macrohard’ AI Venture

• OpenAI’s ChatGPT Agents Take Aim at Excel and PowerPoint

• Amazon Launches AgentCore: Scalable, Secure AI Agents Made Easy

• Google Launches AI Phone Agent, Supercharges Search with Deep Reasoning

• ChatGPT to Take Sales Commissions as OpenAI Launches In-App Shopping

Elon Musks Reveals GROK 4's Future | MacroHard, Tesla Merger, DoD and more…

Elon Musk’s xAI released Grok 4, a powerful AI model trained using 10x more compute than its predecessor. Backed by Musk’s massive data center, Grok 4 powers simulations of human-like agents and may expand into video generation. 

It’s being tested for government use and integrated with Tesla cars. Early results show strong reasoning abilities, and Musk hints at launching a new company—possibly called “Macrohard”—to lead in next-gen AI platforms.

WATCH THE VIDEO ON YOUTUBE

OpenAI’s ChatGPT Agents Take Aim at Excel and PowerPoint

OpenAI is trialing new ChatGPT helpers that let users build and edit slide shows and spreadsheets straight inside the chat window. The files open in PowerPoint and Excel but work without those programs because they use open formats. Future helpers will write reports from business data and even book meetings. Current versions run slowly and sometimes make mistakes. Shared editing is planned. The push could worsen OpenAI’s alliance with Microsoft.

Why this matters

1. New work flows – Turning ChatGPT into a one‑stop office tool could change how millions create documents and manage tasks.
   

2. Platform independence – By relying on open file types, OpenAI reduces dependence on Microsoft and Google, nudging the AI field toward open standards.
   

3. Competitive tension – Expanding ChatGPT’s skills may strain its partnership with Microsoft, reshaping alliances and investment paths in the AI industry.

Amazon Launches AgentCore: Scalable, Secure AI Agents Made Easy

Amazon introduces AgentCore, a set of cloud services that let developers deploy and run AI agents securely and at scale. AgentCore supplies serverless runtime, memory, identity, observability, code interpreter, browser automation, and a gateway for connecting APIs, removing heavy infrastructure work. Developers can mix and match modules, keep favorite frameworks, and publish agents via AWS Marketplace. Preview is free until Sept 16 2025 in four regions; full billing starts Sept 17, after the preview.

Why this matters

1. Faster real‑world use – AgentCore handles hosting, security, and monitoring, so teams can turn demos into working products quickly.
   

2. Open connections – It works with any model and follows shared “agent talk” rules, pushing the field toward tools that play nicely together.
   

3. Cloud competition – AWS now matches rivals on agent hosting, giving users more choice and likely sparking fresh price and feature races in AI.

Google Launches AI Phone Agent, Supercharges Search with Deep Reasoning

Google is rolling out an AI-powered business-calling feature in the U.S., allowing its AI to call local businesses for pricing and availability details on your behalf. It clearly identifies itself as automated, addressing past criticism. Google also upgrades Search’s AI Mode with Gemini 2.5 Pro for advanced reasoning, math, and coding, and adds “Deep Search,” which performs hundreds of queries and compiles comprehensive, cited reports in minutes, competing with ChatGPT Search and Perplexity.

Why this matters

1. Agentic AI in practice – Google brings real-world task automation to consumers, signaling a shift toward everyday AI assistants handling errands.
   

2. Search evolution – Deep Search and Gemini 2.5 Pro transform Google Search into an AI reasoning engine, reshaping how people find and synthesize information.
   

3. Competitive pressure – These updates intensify Google’s battle with OpenAI and Perplexity, accelerating innovation in AI-driven search and personal agents.

🧠RESEARCH

EXAONE 4.0: Unified Large Language Models Integrating Non-reasoning and Reasoning Modes

EXAONE 4.0 is LG’s new AI model that combines fast response (non-reasoning) with deep thinking (reasoning). It supports English, Korean, and now Spanish. It comes in two sizes for different devices and outperforms similar models. It’s designed for real-world AI tasks and is free to use for research on Hugging Face.

SpeakerVid-5M: A Large-Scale High-Quality Dataset for Audio-Visual Dyadic Interactive Human Generation

SpeakerVid-5M is a massive dataset of over 8,700 hours of human video clips designed for training lifelike virtual humans that can talk and interact. It includes varied conversation styles and offers both large-scale and high-quality data. A benchmark system and tools are also provided to support future research and development.

Vision-Language-Vision Auto-Encoder: Scalable Knowledge Distillation from Diffusion Models

The VLV auto-encoder creates powerful image captioning using a smart mix of existing AI models. Instead of needing billions of image-text pairs, it trains mostly on images by distilling knowledge from text-to-image models. It matches top captioning models like GPT-4o, while staying cheap—under $1,000 in training cost.

🛠️TOP TOOLS

Logo Redesign AI - AI-powered tool that transform and enhance your brand’s visual identity with ease and efficiency.

Copyleaks AI - Designed to identify and flag text generated by AI models.

Figurative Language Checker - Scans your text and ultimately identifies figurative language elements like similes, metaphors, and personification.

Neural Frames - Audioreactive AI animations for musicians, creatives and visual artists.

Augie AI - Designed to make video-first marketing accessible and intuitive for businesses of all sizes.

📲SOCIAL MEDIA

🗞️MORE NEWS

• OpenAI plans to earn revenue by taking a commission from product sales made through ChatGPT, integrating a checkout system for transactions within the platform.Partnerships with Shopify and focus on enhancing user experience and personalized responses through memory upgrades are key aspects of this ecommerce venture.

• OpenAI researcher Jason Wei, known for his work on o1 and advanced AI models, is leaving to join Meta’s superintelligence lab. Another researcher, Hyung Won Chung, is also making the move, signaling Meta's growing AI ambitions.

• Two key developers of Anthropic’s coding tool, Claude Code, briefly left to join rival Anysphere—the team behind Cursor—but have now unexpectedly returned to Anthropic, highlighting fierce competition for AI talent in coding tools.

• Scale AI is laying off 14% of its staff after Meta’s $14.3 billion investment and CEO Alexandr Wang’s move to Meta. The company admits it grew too fast and is now restructuring for efficiency.

• Chinese AI startup MiniMax has filed for a Hong Kong IPO aiming for a $4 billion-plus valuation. Backed by Alibaba and Tencent, it plans to go public by year-end, amid growing global interest in generative AI.

• Former Google researchers have built a new AI agent that learns how software is created by studying real company data. This approach may lead to smarter coding assistants and move AI closer to true understanding.

• Researchers developed EchoNext, an AI model that accurately detects structural heart disease from ECG data, outperforming cardiologists in trials. It can expand access to early heart screening and has released public datasets for further research.

• Researchers from OpenAI and Anthropic criticize Elon Musk’s xAI for skipping standard AI safety practices. Grok 4’s troubling behavior and lack of transparency raise concerns, prompting calls for regulation and greater accountability in AI deployment.