Need a personal assistant? We do too, that’s why we use AI.

Ready to embrace a new era of task delegation?

HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.

Learn how to integrate AI into your own processes, allowing you to optimize your time and resources, while maximizing your output with ease.

Get the free guide here.

Today:

• Grok’s iPhone App Goes Live

• SLMs Outperform OpenAI’s o1-preview

• Google Debuts AI Podcast Summaries

• Cohere Introduces North for Enterprises

• Blackstone Invests $300M in AI

Grok’s iPhone App Goes Live

xAI has launched Grok’s iPhone app in the US, allowing users free access to the AI chatbot previously limited to X Premium subscribers. The app, still in beta, can answer questions, generate images, and provide real-time information. Grok competes with popular chatbots like ChatGPT and Google Gemini. A dedicated Grok.com website is also in development. xAI has raised $12 billion, supported by strategic investors including Nvidia and AMD.

SLMs Outperform OpenAI’s o1-preview

Microsoft's rStar-Math technique enhances small language models (SLMs), enabling them to excel in mathematical reasoning tasks. Using Monte Carlo Tree Search and step-by-step reasoning, the method improved smaller models like Qwen-7B to outperform OpenAI's o1-preview on benchmarks. rStar-Math achieved 90% accuracy on MATH and ranked in the top 20% of AIME competitors. This breakthrough highlights SLMs as cost-effective, high-performing alternatives to massive AI systems, reducing computational and environmental demands.

Google Debuts AI Podcast Summaries

Google is testing "Daily Listen," an AI-powered feature in its Discover feed that creates five-minute personalized podcast summaries based on user interests. Available in the U.S. via Search Labs on Android and iOS, episodes are generated daily. While useful, the feature can produce inaccuracies when synthesizing diverse content, making it a helpful but unreliable source. The system shares similarities with Google’s NotebookLM technology for podcasts.

Cohere Introduces North for Enterprises

Cohere has launched North, a secure AI platform designed for privacy-focused enterprises in finance, healthcare, and more. Combining language models, search, and automation, North excels in customized workflows and data security. It outperformed Microsoft Copilot and Google Vertex AI in key areas like IT and finance. Featuring Compass, a powerful search tool, North reduces task times by over 80%, aiming to transform enterprise AI adoption with security and practical customization at its core.

Blackstone Invests $300M in AI

Blackstone is investing $300 million in DDN, a California-based company specializing in data storage and analysis for AI. This marks DDN's first external funding, valuing the company at $5 billion. The move aligns with Blackstone’s strategy of backing AI-driven infrastructure, reflecting growing demand for efficient data solutions in the rapidly evolving artificial intelligence industry.

🧠RESEARCH

rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking

rStar-Math demonstrates that small language models (SLMs) can excel in math reasoning, surpassing OpenAI's o1 model without distillation. Using Monte Carlo Tree Search, innovative training methods, and self-evolution, it achieves state-of-the-art performance on math benchmarks, solving complex problems with verified reasoning. Results rival top high school students in math competitions.

Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Though

The Meta Chain-of-Thought (Meta-CoT) framework enhances traditional reasoning in large language models by modeling the reasoning behind each step. Using process supervision, synthetic data, and search algorithms, it trains models to produce Meta-CoTs. This approach aims to improve AI reasoning, bridging the gap toward more human-like problem-solving capabilities.

URSA: Understanding and Verifying Chain-of-thought Reasoning in Multimodal Mathematics

The URSA framework enhances multimodal mathematical reasoning in large language models by addressing the scarcity of high-quality Chain-of-Thought (CoT) data. Using novel data synthesis and verification strategies, URSA achieves state-of-the-art performance and robust test-time scaling. Its verifier model demonstrates strong generalization, improving reasoning accuracy and out-of-distribution capabilities.

Agent Laboratory: Using LLM Agents as Research Assistants

Agent Laboratory is an autonomous framework leveraging large language models to streamline scientific research. It automates literature reviews, experimentation, and report writing, producing high-quality outputs with human feedback. This approach reduces research costs by 84%, accelerates discovery, and allows researchers to focus on creative ideation rather than tedious tasks.

SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images

SPAR3D introduces a two-stage approach for reconstructing 3D objects from single images. Combining fast point cloud generation with detailed mesh creation, it bridges regression and generative modeling. This method excels in handling occluded regions, allows interactive edits, and achieves state-of-the-art performance with high efficiency, processing images in just 0.7 seconds.

🛠️TOP TOOLS

HeyGen - AI-powered video creation platform that transforms text, voice, and images into lifelike avatar videos.

Perplexity AI Companion - Chrome extension that blends ChatGPT and Google-like functionality, offering users instant answers, summaries, and the ability to ask questions directly from their browser toolbar while browsing the web.

FakeYou - AI-powered online platform that allows users to generate synthetic speech and perform voice conversions using deep learning technology.

VocAI Shulex - Cut costs and increase customer satisfaction with an all-in one solution powered by AI.

Jenni AI - Writing assistant designed to enhance productivity and creativity for students, researchers, and professionals across various writing tasks

📲SOCIAL MEDIA

🗞️MORE NEWS

• Perplexity's AI search engine now integrates with Tripadvisor to provide detailed hotel info, including images, ratings, and highlights. The update enhances user experience and plans to expand to restaurants and experiences.

• Character.AI faces backlash after users created school shooter chatbots, raising ethical concerns about generative AI moderation. The platform removed these chatbots, implemented stricter safeguards, and highlighted the importance of parental oversight for minors.

• Elon Musk claims AI training has exhausted human data, urging a shift to AI-generated synthetic data despite risks like inaccuracies and "model collapse." Experts warn over-reliance could harm creativity and output quality.

• Nvidia CEO Jensen Huang predicts IT departments will evolve into HR for AI agents, emphasizing their critical role in managing AI systems. Huang views AI agents as transformative, presenting vast economic opportunities.

• AI is set to transform Wall Street, potentially cutting 200,000 jobs within five years. While roles in data analysis and operations face displacement, new opportunities will emerge, emphasizing AI oversight and collaboration.