Get Your Free ChatGPT Productivity Bundle

Mindstream brings you 5 essential resources to master ChatGPT at work. This free bundle includes decision flowcharts, prompt templates, and our 2025 guide to AI productivity.

Our team of AI experts has packaged the most actionable ChatGPT hacks that are actually working for top marketers and founders. Save hours each week with these proven workflows.

It's completely free when you subscribe to our daily AI newsletter.

Subscribe to get instant access

Today:

• Apple Develops Chips For Smart Glasses

• Microsoft Hires Meta Exec For Copilot

• Hugging Face Unveils Open Computer Agent

• Google Adds Implicit Caching To Gemini

• Alibaba Launches ZeroSearch For AI Training

Apple Develops Chips For Smart Glasses

Apple is developing specialized chips to power future smart glasses, advanced Macs, and AI servers. The smart glasses chip aims to compete with Meta's Ray-Ban spectacles. The new silicon also targets AI server efficiency, marking a strategic move to bolster Apple’s AI and AR hardware ecosystem. This development underscores Apple's commitment to building in-house silicon to enhance performance and integration across its expanding range of smart and AI-driven devices.

Why This Matters

1. Enhanced AI Hardware: Custom chips for AI servers indicate Apple's push to optimize AI processing, making devices more efficient and integrated.
   

2. Smart Glasses Evolution: The move into AR glasses with proprietary chips could set a new standard for wearable AI, challenging Meta's dominance.
   

3. Ecosystem Expansion: Integrating AI-specific silicon across Apple's products strengthens its AI ecosystem, blending wearable tech with high-performance computing.

Microsoft Hires Meta Exec For Copilot

Microsoft has hired Mark D’Arcy, a former Meta marketing executive, as the creative director for its Copilot AI chatbot. Reporting to AI chief Mustafa Suleyman, D’Arcy will focus on branding and advertising to position Copilot as an empathetic AI companion. This move highlights Microsoft’s effort to compete with OpenAI and Google in the chatbot market, where Copilot’s user base lags behind ChatGPT. D’Arcy's experience aims to boost Copilot’s appeal and user growth.

Why This Matters

1. Strategic Branding for AI Companions: Microsoft is emphasizing Copilot’s emotional intelligence, a key differentiator in the crowded AI chatbot market.
   

2. Experienced Leadership: Hiring a veteran marketer from Meta signals a strategic move to improve Copilot's user engagement and public perception.
   

3. Competitive Positioning: As OpenAI's ChatGPT dominates the market, Microsoft's investment in branding reflects a push to narrow the gap and increase Copilot's adoption.

Hugging Face Unveils Open Computer Agent

Hugging Face unveiled the Open Computer Agent, an open-source AI tool designed to perform basic computer tasks via a web browser. Built on the "smolagents" framework and using Alibaba's Qwen-VL vision model, it can interact with a Linux-based virtual machine and applications like Firefox. However, the tool has significant limitations, including slow responses and frequent errors during simple tasks, making it more of a tech demo than a practical, everyday assistant.

Why This Matters

1. Open-Source AI Agents: Hugging Face’s focus on open-source development encourages community collaboration and innovation in AI agent research.
   

2. Real-World Interaction Challenges: The tool highlights the current technical barriers in creating reliable, human-like AI agents for everyday computer tasks.
   

3. Competitive Landscape: As AI companies race to develop practical digital assistants, this release shows how far the field still needs to progress to match commercial solutions.

🧠RESEARCH

Unified Multimodal Chain-of-Thought Reward Model through Reinforcement Fine-Tuning

UnifiedReward-Think is a multimodal reward model that integrates long chain-of-thought (CoT) reasoning to improve accuracy and robustness in vision tasks. It uses reinforcement fine-tuning with diverse data to enhance reasoning depth. The model outperforms existing methods in visual understanding and generation, offering a more reliable way to align AI outputs with human preferences.

FlexiAct: Towards Flexible Action Control in Heterogeneous Scenarios

FlexiAct is a new method for generating videos where subjects perform actions from reference videos, even if their appearance, pose, or viewpoint differ. It uses a flexible action transfer technique with RefAdapter for spatial adaptation and FAE for motion consistency. FlexiAct outperforms existing methods in preserving identity while allowing diverse action customization.

Unified Multimodal Understanding and Generation Models: Advances, Challenges, and Opportunities

This paper surveys unified models that combine multimodal understanding and image generation. It categorizes current approaches into three types: diffusion-based, autoregressive-based, and hybrid models. The study highlights challenges like integrating diverse architectures and managing cross-modal attention. It also provides resources and benchmarks to aid future research, aiming to inspire progress in unified multimodal AI.

ZeroSearch: Incentivize the Search Capability of LLMs without Searching

ZeroSearch is a new method to train language models for search tasks without using real search engines, saving on costs and reducing noise from unreliable data. It fine-tunes models to generate both relevant and noisy documents, then challenges them with increasingly difficult scenarios. Models trained with ZeroSearch match or surpass real search engines in performance.

RetroInfer: A Vector-Storage Approach for Scalable Long-Context LLM Inference

RetroInfer is a system that speeds up long-context language model inference by treating key-value (KV) caches as vector storage. Using techniques like attention-aware indexing and efficient cache management, it achieves up to 4.5X faster processing within GPU limits and 10.5X faster when using CPU memory, all while maintaining model accuracy.

🛠️TOP TOOLS

PicFinder AI - AI-powered image generation tool that transforms textual descriptions into unique visual content.

Unboring AI - AI-powered online platform that offers innovative tools for face swapping, photo animation, and video restyling, allowing users to create engaging and entertaining content with just a few clicks.

Crayo - Transform simple text inputs into viral-ready videos.

Dubverse AI - AI-powered platform that revolutionizes video dubbing and subtitling, offering creators and businesses the ability to make their content multilingual instantly across 30+ languages with human-like AI voices.

Bing Create - AI-powered tool that transforms text descriptions into visual art, offering users the ability to generate unique images from their ideas. 

📲SOCIAL MEDIA

🗞️MORE NEWS

• Google introduced "implicit caching" in its Gemini API, reducing costs by 75% for repetitive tasks. Unlike manual caching, this new feature automatically reuses common data, saving developers time and money when using Gemini 2.5 models.

• Alibaba's ZeroSearch trains AI to search without using real search engines, cutting training costs by 88%. It uses simulation rather than external APIs, allowing AI to generate relevant documents itself. This approach offers cost savings and greater control for developers.

• Ex-Synapse CEO Sankaet Pathak is raising $100 million for his new humanoid robotics startup, Foundation, at a $1 billion valuation. Despite Synapse's bankruptcy in 2024, Pathak aims to develop advanced robots to tackle labor shortages.

• Google is introducing AI-powered scam protection for Chrome using Gemini Nano, enhancing security against online scams and phishing. The update includes scam warnings on Android and improved scam detection for risky websites and notifications.

• OpenAI now allows developers to fine-tune the o4-mini reasoning model using reinforcement learning (RFT), enabling customization for specific enterprise needs. Early adopters report significant performance improvements, and RFT offers flexible, cost-effective model adaptation through OpenAI’s platform.

• OpenAI is discussing a project called "cderGPT" with the FDA to use AI for drug evaluation. Talks involve FDA AI Officer Jeremy Walsh and Musk-affiliated representatives. No formal deal yet, but the FDA has been exploring AI for years.

• Microsoft has banned employees from using the DeepSeek app due to data security and propaganda concerns, citing the risk of data being stored in China. Despite this, DeepSeek’s R1 model remains available on Microsoft’s Azure cloud service.