AI & Startups December 16th - December 22nd

AI Product News

?? Microsoft Releases Small, Powerful Phi-4

Microsoft has unveiled Phi-4, a 14B parameter language model that rivals—and often surpasses—larger competitors like GPT-4o and Gemini Pro 1.5 in mathematical reasoning and complex problem-solving, demonstrating that size isn’t everything in AI.

The Details:

?? Outperforms Larger Models: Phi-4 excels in math and complex reasoning, surpassing models like Gemini Pro 1.5 and even its own teacher, GPT-4o, on graduate-level STEM Q&A and math competition benchmarks.

?? Synthetic Data Training: The model was trained on approximately 400B tokens of AI-generated, high-quality synthetic data, validated for accuracy and relevance.

?? Extended Input Capacity: Phi-4 features an upgraded mechanism allowing it to process inputs of up to 4,000 tokens, doubling the capacity of its predecessor, Phi-3.

?? Research Preview: Phi-4 is currently available for a limited research preview on Azure AI Foundry, with a broader release planned for Hugging Face.

Why It Matters: Microsoft's Phi-4 challenges the notion that larger models are inherently superior. By emphasizing smarter architectures and innovative training approaches, it proves that smaller, more efficient models can rival or exceed the capabilities of massive competitors. This shift could redefine the future of AI development, focusing on optimization and specialization rather than sheer scale.

??? ChatGPT Gains ‘Projects’ for Chat Organization

OpenAI has introduced Projects for ChatGPT as part of its ‘12 Days of OpenAI’ event, offering a new way to organize conversations, files, and instructions into cohesive workspaces.

The Details:

?? Project-Specific Folders: Users can now bundle related chats, documents, and custom instructions into individual workspaces for seamless organization and shared context.

?? Enhanced Capabilities: Each Project utilizes GPT-4o while retaining core features like Canvas, DALL-E, and web search.

?? Subscriber Rollout: Currently available to Plus, Pro, and Teams users, with Enterprise and Education access planned for January.

?? Cross-Platform Availability: Projects can be created on the web and Windows apps, with mobile and Mac users gaining view and chat functionality for existing Projects.

Why It Matters: While not entirely novel (Anthropic added Projects for Claude in June), this update addresses a critical workflow need. Projects eliminate the hassle of resetting context and re-entering instructions when switching tasks, making ChatGPT more practical for long-term and multi-threaded use.

?? Google Launches Next-Gen Video and Image Models

Google has unveiled Veo 2 and Imagen 3, setting new benchmarks in AI-generated video and image quality. These models showcase remarkable advancements in realism, detail, and prompt adherence, solidifying Google’s dominance in generative AI.

Veo 2:

?? High-Resolution Clips: Generates 8-second videos in 4K resolution (720p available at launch), with cinematic-grade control and realism.

?? Improved Physics and Detail: Excels in realistic movement and physics simulations, significantly reducing visual artifacts and hallucinations.

?? Human Evaluations Leader: Outperformed competitors, including OpenAI’s Sora, in prompt adherence and visual quality.

?? Gradual Rollout: Available via the VideoFX waitlist, with YouTube Shorts integration planned for 2025.

Imagen 3:

?? Enhanced Artistic Quality: Offers vibrant colors, improved texture details, and seamless composition across diverse artistic styles.

??? Accurate Prompt Interpretation: Delivers on complex user prompts with unmatched precision, handling intricate scenes and text rendering effortlessly.

?? Top Human Evaluations: Outshined Midjourney, Flux, and Ideogram in preference, visual quality, and adherence to prompts.

?? Global Availability: Now accessible via Google Labs’ ImageFX, rolling out to 100+ countries.

Why It Matters: With Veo 2 and Imagen 3, Google is ending 2024 on a high note, building on the success of Gemini 2.0. These models redefine the standard for generative video and image AI, pushing past the competition and demonstrating Google’s leadership in AI innovation. While OpenAI garners holiday buzz, Google is delivering cutting-edge results.

?? ChatGPT Search Goes Free for Everyone

OpenAI has expanded its ChatGPT Search feature, making it free for all users and introducing voice search capabilities and enhanced mobile experiences as part of Day 8 of the company’s livestream event.

The Details:

?? Universal Access: Previously exclusive to premium users, Search is now free for all logged-in users, accessible via a globe icon with faster response times.

??? Voice Search Integration: Premium users in Advanced Voice Mode can now conduct searches through natural spoken prompts, blending web access with conversational AI.

?? Revamped Mobile Experience: Features include enhanced visual layouts for local businesses and seamless integration with Google and Apple Maps.

?? Default Search Option: Users can set ChatGPT Search as their default search engine, displaying relevant links before ChatGPT’s AI responses for faster results.

?? What’s Next: OpenAI teased a ‘mini Dev Day’ announcement, hinting at upcoming features.

Why It Matters: ChatGPT’s web search capability is a pivotal step toward creating a fully agentic assistant, particularly with Advanced Voice Mode making the tool a smarter, more intuitive alternative to Siri or other voice assistants. As AI reshapes the way we interact with search engines, ChatGPT Search is poised to lead the next big shift in the AI era.

?? The First Generative AI Physics Simulator

A team of researchers from 20 labs has introduced Genesis, an open-source physics engine that blends generative AI with ultra-fast simulations. This groundbreaking platform is poised to transform how AI learns to navigate and interact with the physical world.

The Details:

? Unprecedented Speed: Genesis achieves 43 million FPS on a single RTX 4090 GPU, running 430,000 times faster than real-time physics.

?? Pure Python Build: The engine is 10-80x faster than existing solutions like Isaac Gym and MJX, offering unmatched performance with simplicity.

?? Real-World Training: Trains robot locomotion policies transferable to real-world applications in just 26 seconds.

??? Generative 4D Environments: Open-source platform will soon include a framework to create complex environments from simple prompts.

Why It Matters: Genesis pushes the boundaries of physical AI, allowing machines to run millions of simulations at speeds never seen before. Its open-source nature and generative capabilities make it a catalyst for innovation, accelerating progress in robotics, AI training, and physical environment modeling. This technology could redefine how AI understands and interacts with the real world.

? Google Releases Experimental 'Reasoning' AI

Google has unveiled Gemini 2.0 Flash Thinking Experimental, a free-to-use AI model designed to tackle complex problems by "thinking" through them, offering faster performance and explicit reasoning capabilities.

The Details:

?? Reasoning in Action: Gemini 2.0 Flash explicitly shows its thought process while solving problems, similar to OpenAI’s o1 model.

? Speed Advantage: Early users report significantly faster performance compared to competing reasoning models, even while handling complex tasks.

?? Dynamic Computation Time: The model adjusts computation time to improve reasoning accuracy, resulting in longer but more precise responses.

?? Top-Ranked AI: Now ranked #1 across all categories on Chatbot Arena, Gemini 2.0 Flash is freely accessible through AI Studio, the Gemini API, and Vertex AI.

Why It Matters: Google’s Gemini 2.0 Flash Thinking Experimental underscores the shift in AI development towards smarter and more efficient models. While competitors like OpenAI increase pricing for advanced reasoning capabilities, Google’s decision to make cutting-edge AI freely available could democratize access and drive faster adoption. This move cements Google’s position as a leader in AI innovation, challenging the status quo.

?? GitHub Copilot Goes Freemium

GitHub, owned by 微软 , has launched a free tier of its popular AI coding tool, Copilot, available through VS Code. This move coincides with GitHub celebrating a significant milestone of 150M developers on its platform, up from 100M in early 2023.

The Details:

?? Freemium Access: The new free tier offers 2,000 monthly code completions and 50 chat messages, fully integrated into VS Code and GitHub's dashboard.

?? AI Model Options: Users can choose between Anthropic’s Claude 3.5 Sonnet or OpenAI’s GPT-4o, while premium models like o1 and Gemini 1.5 Pro remain part of paid plans.

??? Robust Features: Includes multi-file editing, terminal assistance, and project-wide context awareness for intelligent code suggestions.

?? Developer Milestone: GitHub has reached 150M developers worldwide, with plans to grow that number to 1B in the future.

Why It Matters: GitHub’s move to make Copilot more accessible reflects a broader trend of AI tools evolving into essential utilities rather than premium add-ons. By lowering price barriers, GitHub is positioning itself as the go-to platform for developers while staying competitive against free AI coding alternatives. This freemium shift is a significant step toward democratizing AI-assisted coding for the global developer community.

?? NVIDIA ’s Affordable, Palm-Sized AI Supercomputer

Nvidia has unveiled the Jetson Orin Nano Super Developer Kit, a $249 generative AI supercomputer that offers substantial performance gains in a compact, palm-sized form factor—at half the cost of its predecessor.

The Details:

? Enhanced Performance: Delivers 1.7x the performance, 70% more processing power, and a 50% memory boost compared to the previous model.

?? Multi-Task AI Capability: Simultaneously powers chatbots, controls robots, and processes visual data from multiple cameras with ease.

??? Comprehensive Ecosystem: Supports AI frameworks and tools like Isaac for robotics and Metropolis for vision AI through Nvidia’s software platform.

?? Free Upgrades: Existing Jetson Orin Nano users can achieve the same 1.7x performance boost via a free software update.

Why It Matters: Nvidia’s Jetson Orin Nano Super Developer Kit could democratize advanced AI development, much like the Raspberry Pi did for DIY computing. Its affordability and versatility make it accessible to developers and hobbyists, sparking a new wave of innovation in areas like robotics, AI-powered devices, and creative tools. With barriers to entry lower than ever, the next big AI breakthroughs might come from garages and dorm rooms.

?? Capital Watch

?? AI Firm Databricks Secures $10 Billion Series J Investment at $62 Billion Valuation

Databricks has raised $10 billion in its Series J funding round, one of the largest VC rounds in history, valuing the data and AI firm at $62 billion.

The Details:

?? Massive Investment: The $10B Series J round, led by Thrive Capital , Andreessen Horowitz , DST Global , and others, included participation from new and existing investors such as Ontario Teachers' Pension Plan , ICONIQ Growth , and Wellington Management, Inc. .

?? Rapid Growth: Databricks grew revenue by over 60% year-over-year, achieving a $600M run rate for Databricks SQL and expecting to cross $3B revenue by Q4 FY25.

?? International Expansion: Funds will support new AI products, acquisitions, and global operations, with India serving as a key market, home to over 400 employees and a major R&D hub in Bengaluru.

?? Platform Power: The Data Intelligence Platform democratizes access to AI and data analytics, helping organizations innovate, increase revenue, lower costs, and reduce risk.

?? Profitability Milestone: Q3 FY25 marks the company’s first quarter with positive free cash flow, positioning Databricks for sustainable growth.

?? AI Startup Perplexity Closes Funding Round at $9 Billion Value

AI search startup Perplexity has secured a $500M funding round, tripling its valuation to $9B as it takes bold steps to challenge traditional search engines.

The Details:

?? Valuation Surge: Perplexity’s valuation soared from $1B in April to $9B in this round, despite facing lawsuits from major publishers.

?? Growing User Base: Since its 2022 launch, the startup has attracted over 15M active users, bolstered by features like one-click shopping and financial analysis.

?? Publisher Partnerships: Revenue-sharing deals with publishers such as Time and Fortune aim to address content usage concerns.

?? Strategic Acquisition: The acquisition of Carbon enables seamless integration with platforms like Notion and Google Docs.

Why It Matters: Perplexity’s meteoric rise showcases the AI-driven disruption reshaping the search industry. As traditional players like Google adapt to AI-era demands, Perplexity’s innovations and strategic moves could define whether it stays ahead in this high-stakes race.

?? AI Cloud Startup Vultr Raises $333M at $3.5B Valuation

Vultr, an AI cloud infrastructure startup, has raised $333M in growth financing, marking its first outside funding round in its 10-year history. The round values the company at $3.5B and highlights the growing competition among chipmakers to dominate the AI infrastructure market.

The Details:

?? Valuation Milestone: Vultr’s first external funding round brings its valuation to $3.5B, positioning it as a major player in the cloud infrastructure space.

?? Lead Investors: The round was co-led by AMD Ventures and LuminArx Capital Management, reflecting AMD's commitment to challenging Nvidia in the AI infrastructure domain.

??? Capital Allocation: Vultr plans to use the funds to acquire more GPUs, critical for powering large language models and scaling AI services.

?? AMD Collaboration: With AMD Instinct accelerators, Vultr strengthens its reputation as a secure, scalable cloud solutions provider, empowering enterprises and AI innovators.

?? Broader AMD Investments: This funding follows AMD Ventures' recent AI-focused investments, including Celestial AI, Scale AI, Cohere, Liquid AI, and Ayar Labs.

Why It Matters: The Vultr funding underscores the intensifying race among chipmakers like AMD, Nvidia, and Intel to capture the AI infrastructure market. By partnering with an up-and-coming cloud provider like Vultr, AMD aims to boost its market visibility and compete more effectively against Nvidia, a dominant force in the sector. This strategic investment signals a transformative era in AI cloud infrastructure.

?? Quantum AI Startup SandboxAQ Raises $300M at $5.6B Valuation

SandboxAQ, an AI and quantum computing startup spun off from Alphabet Inc. , has raised over $300M in new funding, bringing its valuation to $5.6B. The funding aims to accelerate the development of large quantitative models (LQMs) and AI applications across multiple industries.

The Details:

?? Massive Funding: The round, led by Fred Alger Management, T. Rowe Price, and Breyer Capital, follows a $500M raise last year for its quantum computing platform.

?? LQM Focus: Unlike generative AI models like ChatGPT, SandboxAQ focuses on numerical data, training large quantitative models to address quantitative relationships in industries like healthcare and finance.

?? Industry Applications: Capital will be invested in modules for enterprise use cases, including drug discovery, materials science, cybersecurity, and encryption.

?? Hardware Efficiency: SandboxAQ’s AI models are optimized to train on existing hardware like Nvidia GPUs and are positioned to leverage future advancements in quantum chips.

???? Leadership: Led by CEO Jack Hidary, the company has former Google CEO Eric Schmidt as chairman, highlighting its strong leadership and vision.

Why It Matters: SandboxAQ’s unique approach to AI and quantum computing positions it as a transformative player in industries reliant on quantitative analysis. Its ability to bridge current AI capabilities with emerging quantum technologies could reshape sectors like life sciences, finance, and cybersecurity, cementing its role as a pioneer in the next wave of technological innovation.

?? Chinese AI Startup Zhipu AI Raises $412M in Funding

Beijing-based AI startup Zhipu has secured 3 billion yuan (US$412M) in its latest funding round, marking another major milestone following its US$400M raise in May.

The Details:

?? Significant Backing: Zhipu’s funding highlights continued investor confidence, with past support from Alibaba, Tencent, and Saudi Aramco’s venture arm.

?? Revenue Growth: The startup reported a notable increase in commercial revenue for 2024, reflecting its expanding market impact.

?? Valuation Milestone: The May funding round valued Zhipu at approximately US$3 billion, underlining its rapid growth in the AI space.

Why It Matters: Zhipu’s success demonstrates the growing prominence of Chinese AI startups in the global market. With backing from tech giants and international investors, Zhipu is well-positioned to further innovate and compete in the evolving AI landscape.

?? AI Startup Basis Raises $34 Million for Accounting Automation

Basis, an AI-powered accounting automation startup, has raised $34M in a Series A funding round led by Khosla Ventures . The funding will support its mission to address the accounting talent shortage with its autonomous AI agent.

The Details:

?? Significant Backing: Investors include Khosla Ventures, NFDG (led by Nat Friedman and Daniel Gross), OpenAI board members Larry Summers and Adam D'Angelo, and Google’s chief scientist Jeff Dean.

?? AI Agent for Accounting: Basis’ product acts as a junior accountant, automating workflows like transaction entry and data accuracy checks, while integrating with popular tools like QuickBooks and Xero.

?? Time Reduction: Accounting firms like Wiss report up to 30% time savings, allowing accountants to focus on reviewing AI-generated outputs rather than manual data entry.

?? Talent Shortage Solution: Basis addresses a growing shortage of accountants, with fewer candidates sitting for the CPA exam as baby boomers retire.

?? Industry Vulnerability: Accounting is among the professions most impacted by AI. A 2023 OpenAI report concluded that AI-driven automation could affect 100% of accountants' tasks.

Why It Matters: Basis is tackling one of the accounting industry's biggest challenges: labor shortages and rising workloads. By introducing AI agents, it not only automates time-consuming tasks but also redefines roles, positioning the profession for a future where humans collaborate with AI to deliver higher-value insights.

?? Indian WealthTech AI Startup Mili Raises $2M Seed Funding

Mili, an AI-powered meeting assistant for wealth management firms, has raised $2M in a seed funding round led by Chiratae Ventures and BoldCap . The funding will drive product enhancements and support the platform's mission to streamline workflows for wealth advisors.

The Details:

?? Seed Round Investors: Backed by Chiratae, BoldCap, Sparrow Capital, SFMG Wealth Advisors, Gregg Fisher (Quent Capital), and Better Capital.

??? Product Enhancements: Funds will focus on expanding customization features, improving platform integrations, and strengthening engineering and customer success teams.

?? Value Proposition: Mili reduces administrative workloads, streamlines workflows, and ensures compliance for wealth advisors, enabling them to focus on client needs.

Why It Matters: Mili addresses a critical pain point in wealth management by leveraging AI to optimize operations and compliance. Its ability to reduce administrative burdens makes it a valuable tool for advisors, positioning it for growth in the rapidly evolving WealthTech space.