There’s a reason 400,000 professionals read this daily.

Join The AI Report, trusted by 400,000+ professionals at Google, Microsoft, and OpenAI. Get daily insights, tools, and strategies to master practical AI skills that drive results.

Sign up now for free and work smarter, not harder.

Today:

• OpenAI Moves Funding to SoftBank

• DeepSeek Considers First Outside Funding

• Meta Opens Bengaluru AI Hub

• DeepMind Sets Pricing for Veo 2

• X Introduces AI Ad Generation

OpenAI Moves Funding to SoftBank

OpenAI is shifting its reliance from Microsoft to SoftBank, which will fund most of its AI data centers through the Stargate project. The company aims to raise $40 billion, with SoftBank contributing $30 billion. OpenAI forecasts revenue to triple in 2025, but cash burn will rise to $7 billion. By 2030, computing power will heavily depend on Stargate. OpenAI expects to turn profitable after its massive infrastructure buildout.

DeepSeek Considers First Outside Funding

Chinese AI startup DeepSeek is considering raising outside funding for the first time to support its rapid growth. Despite previously rejecting investors, rising demand for AI chips and servers has drawn interest from Alibaba and state-backed funds. DeepSeek's chatbot app has surged in popularity, but infrastructure challenges persist. U.S. scrutiny complicates its funding decisions, while the company explores using Southeast Asian data centers to bypass chip restrictions and sustain its AI advancements.

Meta Opens Bengaluru AI Hub

Meta is expanding its presence in India with a new office in Bengaluru, hiring engineers and AI specialists. This move follows tech giants like Microsoft, Google, and Amazon, which have strong teams in the region. The Bengaluru office will focus on AI infrastructure, data centers, and internal productivity tools. India, Meta’s largest user base, plays a crucial role in its global strategy, with over a billion people using its platforms.

DeepMind Sets Pricing for Veo 2

Google DeepMind has set pricing for its Veo 2 AI video generation model, charging $0.50 per second via its Vertex cloud platform—$1,800 per hour. While more expensive than OpenAI’s Sora ($200/month with no cap), it remains far cheaper than traditional filmmaking, which costs around $32,000 per second. The cost only covers AI generation, with additional expenses for human labor and multiple iterations to refine the final output.

X Introduces AI Ad Generation

X (formerly Twitter) is launching AI-powered advertising tools to automate ad creation and performance analysis. Using its AI assistant, Grok, the platform can generate ad copy, imagery, and insights with minimal human input. One key feature, “Prefill with Grok,” allows advertisers to input a website URL, letting AI create ready-to-use ads. This move aims to recover lost ad revenue and streamline campaign development for advertisers.

🧠RESEARCH

MLGym: A New Framework and Benchmark for Advancing AI Research Agents

MLGym introduces a new framework and benchmark for testing AI research agents on machine learning tasks. It evaluates large language models in areas like computer vision and reinforcement learning. While models improve on baseline results, they lack true innovation. The open-source framework supports adding tasks and refining AI research capabilities.

S^2R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning

S²R is a framework that improves LLM reasoning by teaching models to self-verify and self-correct. Using supervised fine-tuning and reinforcement learning, it enhances accuracy with minimal data. Tests show significant performance boosts, even for smaller models. The approach requires fewer resources and outperforms traditional training methods.

How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM?

This study examines how much new knowledge can be added to an LLM using LoRA without degrading existing knowledge. Results show that mixing new and known facts yields the best performance, but fine-tuning can still harm external benchmark accuracy. Biases in training data may also skew model responses.

Logic-RL: Unleashing LLM Reasoning with Rule-Based Reinforcement Learning

Logic-RL enhances LLM reasoning using rule-based reinforcement learning. By training on synthetic logic puzzles, the model develops skills like reflection and verification. Key improvements include a structured prompt, strict reward system, and stable training method. With only 5K logic problems, the model generalizes well to advanced math benchmarks.

🛠️TOP TOOLS

Dzine AI - Generate and edit in one place with AI-Powered design workflow.

My Vocal AI - Clone Your Voice to Sing, Speak, and Beyond.

HotBot AI - Ask AI questions for expert bot assistance

Looka AI - AI-powered platform to design a logo and brand you love.

GPTExcel - AI-Powered Spreadsheet Automation:Formulas, Charts & Data Insights.

📲SOCIAL MEDIA

🗞️MORE NEWS

• OpenAI works to prevent AI misuse by authoritarian regimes and malicious actors. Their latest report highlights disrupted threats, supporting global efforts to ensure AI benefits humanity while protecting against exploitation, cyber threats, and covert influence operations.

• Qatar signed a five-year deal with Scale AI to enhance government services using AI-driven tools. The partnership aims to streamline operations through automation, predictive analytics, and advanced data analysis, reinforcing Qatar’s AI leadership in the region.

• Microsoft's Magma AI enhances robots' ability to see, understand, and interact with digital and physical environments. Using multimodal learning, it improves automation, navigation, and real-world actions, advancing AI-driven robotics for broader, more intuitive applications.

• Norwegian robotics firm 1X unveiled Neo Gamma, a humanoid home robot designed for household tasks. With a soft, safety-focused design and AI-powered improvements, it moves toward practical home use but remains in early testing stages.

• Grok 3, xAI’s latest AI model, briefly censored unflattering mentions of Donald Trump and Elon Musk. While the issue was patched, concerns remain over its political neutrality and AI’s role in content moderation.

• AI search startup Genspark raised $100 million in Series A funding, valuing it at $530 million. Competing with Google, it develops AI-powered search features, leveraging multiple models for in-depth online research.

• Researchers identified "temporal heads" in AI models that process time-specific knowledge, enabling updates without full retraining. This discovery could reduce costs while improving AI’s ability to handle evolving information accurately.