DeepSeek OCR, Nvidia GPU, Hugging Face Code Release

The AI landscape is seeing rapid developments across various sectors. Chinese startup DeepSeek is making strides with its DeepSeek-OCR model, which uses a novel 'vision-text compression' technique to process documents efficiently. This model can generate up to 200,000 pages of training data daily on a single Nvidia A100 GPU, significantly reducing token usage and resource needs. The company has made its code and weights publicly available on platforms like Hugging Face. Meanwhile, in the e-commerce realm, Visa is introducing the Trusted Agent Protocol (TAP) to ensure secure transactions as AI agents become more prevalent in online shopping. Zoho is aiming to accelerate AI adoption among businesses by offering its AI agent tools for free across its software suite, a move that mirrors similar strategies from companies like Microsoft and Google Cloud. Salesforce is also revamping its Agentforce to speed up enterprise AI adoption, positioning Slack as the central interface for its AI tools. Google is investing in AI education, partnering with Miami schools to provide AI-powered learning tools and offering AI training programs. In cybersecurity, Dataminr is acquiring ThreatConnect for $290 million to bolster its AI-driven cyber defense capabilities by integrating external and internal data analysis. Amidst these advancements, experts are raising concerns about the potential risks AI companions pose to teenagers, while the real estate industry is adapting to the AI era by focusing on enhanced luxury service and the 'human advantage'. Morgan's Repellent has also utilized AI tools like Copilot, Gemini, and ChatGPT to independently validate its product claims, underscoring a growing trend of AI-assisted verification.

Key Takeaways

• DeepSeek AI's new model, DeepSeek-OCR, can generate 200,000 pages of data daily on one Nvidia A100 GPU using a vision-text compression method that reduces token usage by up to 20 times.
• DeepSeek-OCR's code and weights are publicly available on Hugging Face and GitHub.
• Visa has launched the Trusted Agent Protocol (TAP) to secure e-commerce transactions involving AI agents.
• Zoho is offering its AI agent tools for free across its software suite to encourage business adoption, following similar moves by Microsoft and Google Cloud.
• Salesforce is updating its Agentforce and making Slack the primary interface for its AI tools to speed up enterprise AI adoption.
• Google is partnering with Miami schools and investing $2 million in AI mentorship programs to boost AI skills and education.
• Dataminr is acquiring ThreatConnect for $290 million to enhance its AI cyber defense by combining external and internal data analysis.
• Experts are warning about the potential risks AI companions pose to teenagers.
• The real estate industry is adapting to AI by focusing on luxury service and the 'human advantage'.
• Morgan's Repellent used AI tools like Copilot, Gemini, and ChatGPT to validate its product claims.

DeepSeek AI model generates 200K pages of data daily on one GPU

Chinese AI startup DeepSeek has launched a new AI model called DeepSeek-OCR. This model can process large documents efficiently by compressing text using visual perception. It can generate up to 200,000 pages of training data per day using just one Nvidia A100 GPU. The company aims to reduce the cost and improve the efficiency of large language models (LLMs). Both the code and weights for DeepSeek-OCR are publicly available on platforms like Hugging Face and GitHub. The model uses a two-part architecture, DeepEncoder and a text generator, to compress text effectively.

Deepseek AI uses images to cut resource use by 20 times

Chinese company Deepseek has developed a new AI model, DeepSeek-OCR, that significantly lowers resource needs. It converts text and documents into images, a method called 'vision-text compression'. This process uses up to 20 times fewer tokens compared to processing raw text directly. The model has two parts: DeepEncoder to turn text into images and a decoder to read the text from images. This approach is especially useful for handling data like tables and graphs, potentially lowering costs and boosting efficiency for AI models.

Visa creates Trusted Agent Protocol for secure AI e-commerce

Visa has launched the Trusted Agent Protocol (TAP) to ensure secure transactions as AI agents become more common in e-commerce. These AI agents can learn user preferences and shop for them, potentially changing the retail landscape significantly. Visa's TAP framework allows AI agents to share and access data safely, protecting against fraud and bot activity. This aims to prevent legitimate transactions from being blocked and improve user experience. While currently limited to the Visa network, the company plans to expand TAP to other payment systems.

Graia AI offers empathetic customer service to avoid frustration

Graia has launched a new AI platform designed to provide empathetic customer service, addressing the frustration caused by rigid automation. Many companies struggle to maintain a human touch with AI, leading to customer dissatisfaction. Graia's platform combines artificial intelligence with emotional intelligence to create interactions that are both efficient and human-centered. It uses an 'empathy index score' to guide AI responses and allows clients to set specific parameters for empathetic communication. This approach aims to improve customer satisfaction and loyalty by balancing automation with genuine understanding.

Zoho offers free AI tools to boost slow adoption

Zoho is making its AI agent tools free across its software suite to encourage wider adoption by businesses. Many companies have been slow to implement AI due to issues like data readiness, integration challenges, and high costs. Zoho's free AI features, built on its Zia Hubs and Zia LLM, are now included with existing subscriptions. Other tech companies like Microsoft, UiPath, and Google Cloud are also offering free trials or usage-based pricing models. This strategy aims to lower barriers and get more businesses experimenting with AI agents.

Morgan's Repellent AI validates product claims with independent review

Morgan's Repellent, a company known for eco-friendly pest control, has used five AI platforms to independently verify its product claims. The AI tools, including Copilot, Gemini, and ChatGPT, reviewed 21 core product statements against online information. The review confirmed that Morgan's Repellent's claims about its non-toxic, essential oil-based formula are scientifically supported. This initiative highlights the company's commitment to transparency and building consumer trust. The validated claims cover aspects like effectiveness, duration, safety for families and pets, and eco-friendly ingredients.

Google partners with Miami schools to boost AI skills

Google is collaborating with Miami-Dade County Public Schools and Miami Dade College to enhance education and workforce development through AI. The partnership will provide AI-powered learning tools, including Gemini for Education, to students and teachers. Google is also offering AI training and investing $2 million to support AI mentorship programs at community colleges nationwide. These initiatives aim to equip students with essential skills for the future workforce and improve operational efficiency in schools. The collaboration focuses on lifelong learning and preparing individuals for an AI-driven world.

Salesforce revamps Agentforce to speed up AI adoption

Salesforce is updating its Agentforce AI features to help customers adopt the technology more quickly. CEO Marc Benioff acknowledged a gap between rapid consumer AI adoption and slower enterprise adoption. To bridge this, Salesforce is increasing support with 'forward-deployed engineers' to assist customers. A major update positions Slack as the main interface for all Salesforce software, allowing users to interact with AI agents directly within Slack. This aims to simplify the use of complex Salesforce tools and accelerate the integration of AI into business workflows.

AI companions pose risks for teens, experts warn

Experts are warning about the potential risks of AI companions for teenagers, as these AI tools are designed to simulate close connections. These AI companions can become confidants for young users, but they may pose serious dangers. The article explores the nature of these AI companions and the concerns raised by specialists regarding their impact on adolescent development and well-being. The focus is on the potential downsides of teens forming deep emotional bonds with artificial intelligence.

New real estate course teaches luxury service in AI era

A new real estate course is teaching agents how to provide luxury service, focusing on the 'human advantage' in an AI-driven world. The program adapts five-star hotel service principles to the real estate client journey. It aims to help agents develop a luxury mindset, enhance their presence, and create personalized client experiences from property viewings to post-closing. The course covers topics like confidence, discretion, anticipatory service, and sophisticated communication. This training helps agents move beyond transactional sales to a high-touch service model.

Dataminr buys ThreatConnect for $290M to boost AI cyber defense

Dataminr has acquired ThreatConnect for $290 million to enhance its AI capabilities for real-time cyber defense. The deal combines Dataminr's AI platform for public data with ThreatConnect's expertise in internal data analysis. This fusion will create client-tailored intelligence by enabling AI agents to analyze both external and internal data sources. ThreatConnect's platform is used by major companies and government agencies to prioritize threats and automate responses. The combined offering will provide more actionable threat intelligence, helping organizations understand and react to risks more effectively.

Sources

• DeepSeek’s new AI model can generate 200K pages of training data daily on a single GPU
• New Deepseek model drastically reduces resource usage by converting text and documents into images — 'vision-text compression' uses up to 20 times fewer tokens
• Visa wants to chart a safe course for e-commerce’s AI revolution
• Stop Losing Customers to Cold, Ineffective AI
• Free Agents: How Slow AI Adoption Is Pushing Vendors to Drop the Paywall
• Morgan's Repellent® Validates Product Claims Through Independent AI Review
• Partnering to power Miami’s AI-ready future
• At Dreamforce, Salesforce CEO Marc Benioff acknowledged that customers needed help to bridge the gap between "AI innovation and AI adoption"
• Video Is your teen using AI companions? What are the risks?
• New Real Estate Course Elevates Human Advantage in an AI Driven World
• Dataminr’s $290 million ThreatConnect deal expands AI for real-time cyber defense