GSV's AI News & Updates (04/04/25)
OpenAI raises $40B, Google Gemini 2.5, Runway Gen-4, Apple AI Health Coach, Anthropic "AI Microscope", Uplimit Learning Agents, Ropes $3.1 Seed, Brisk $15M Series A, SchoolAI $25M Series A
Counting down to the AI SHOW and ASU+GSV in San Diego — see you all in just a few days!
General 🚀
OpenAI:
OpenAI Raises $40B in Record-Breaking Round, Valuation Hits $300B: The largest ever for a private tech company — led by SoftBank ($30B) with support from Microsoft, Coatue, Altimeter, and Thrive. $18B is earmarked for its Stargate compute venture, but SoftBank may scale back funding unless OpenAI restructures into a for-profit entity by year-end.
OpenAI Will Release an “Open” Language Model for the First Time Since GPT-2: The model is expected to have reasoning capabilities similar to o3-mini and will undergo rigorous safety evaluation prior to release. This marks a strategic pivot for OpenAI amid rising pressure from Meta’s Llama and China’s DeepSeek.
ChatGPT Gets Major Visual Upgrade with GPT-4o: GPT-4o now powers native image generation in ChatGPT and Sora, enabling users to create and edit images directly in the app.
ChatGPT’s New Image Generator Hits 700M Images in One Week. ChatGPT has reached 20M paid subscribers, generating an estimated $415M in monthly revenue
Google:
Google’s New Gemini 2.5 Pro Quietly Outperforms the Competition: Google’s latest model, Gemini 2.5 Pro, is posting top-tier scores across math, coding, and long-context benchmarks. Despite its GPT-4.5-level performance and free access, Gemini 2.5 Pro is flying under the radar, and its release has been overshadowed by viral Ghibli-style image generation.
Google Gemini is shaking up its AI leadership ranks: Google replaced the head of its Gemini chatbot team, with Josh Woodward (of NotebookLM fame) taking over from Sissie Hsiao. The move signals a shift in focus from core models to product-led AI innovation.
Musk Merges X and xAI in All-Stock Deal: Elon Musk folded social platform X into his AI startup xAI in a $45B all-stock transaction, raising questions about transparency, debt maneuvering, and conflicts of interest. The move gives xAI access to X’s user base and data.
Amazon Debuts Nova Act, Its First General-Purpose AI Agent: Built by a new AGI lab led by ex-OpenAI talent, Nova Act aims to outperform rivals in everyday web tasks — like ordering food or booking reservations — with early benchmarks showing strong results.
Apple Plans AI Health Coach in Biggest Medical Push Yet: Apple is launching its most ambitious health effort yet: an AI doctor-like agent offering personalized coaching, food tracking, and fitness feedback.
China’s Manus Turns Its AI Agent Into a $39 Subscription: Beijing’s Manus AI agent is now a subscription product with tiers at $39 and $199/month, marking one of the fastest monetization moves for a general-purpose agent.
Runway Launches Gen-4, Its Most Advanced AI Video Model Yet: Runway’s new Gen-4 model generates high-fidelity, dynamic videos with consistent characters and world environments — all without fine-tuning. The launch comes amid legal challenges over training data and a rumored $4B fundraise.
Education and the Future of Work 📚
Anthropic Debuts Claude for Education: Anthropic has launched Claude for Education, a new tier aimed at colleges and universities that includes a “Learning Mode” to foster critical thinking. With partners like Northeastern and Instructure, the company is pushing deeper into higher ed — and into direct competition with OpenAI’s ChatGPT Edu.
OpenAI and Anthropic are fighting over college students with free AI: The AI startups are competing for who becomes the next generation’s default AI tool.
Learnosity Research: Teacher Burnout & Grading Workload: 1 in 3 U.S. teachers considered leaving the profession in the past year due to grading workload. 75% of teachers are open to AI tools that could cut grading time in half.
Cengage Report: GenAI Adoption Rising in K12 and Higher Ed: New data from Cengage shows growing optimism and adoption of GenAI in both K12 and Higher Ed. GenAI use in K12 jumped 12% YoY, with 63% of teachers now using it—surpassing HED’s 49% adoption.
Uplimit Launches AI Learning Agents to Scale Corporate Upskilling: Uplimit unveiled a suite of AI agents—focused on skill-building, program management, and teaching assistance—to help enterprises train thousands of employees at once. Customers like Databricks and Procore report >90% completion rates and massive time savings.
Glean Partners with Internet2 to Bring AI Agents to Campuses: Glean has joined Internet2’s trusted consortium to bring its Work AI platform to 500+ higher ed and research institutions. The partnership enables AI-powered search, assistants, and agents to streamline operations, enhance learning, and support campus-wide access to knowledge.
‘Don’t study coding now,’ says Replit CEO: Instead of learning how to code, "learn how to think, learn how to break down problems, learn how to communicate clearly."
The 500 Million Worker Problem: A report warns that India's long-touted demographic advantage, with 500 million youth set to enter the workforce, may collapse under the weight of AI.
Job hunting and hiring in the age of AI: Where did all the humans go?: As companies rely on bots to filter candidates, some applicants are fighting back with AI-generated résumés and deepfake avatars. Recruiters, in turn, are overwhelmed, struggling to spot real humans among the noise.
Startups and Tools 🛠️
Ropes Raises $3.1M to Reinvent Technical Hiring with AI: Ropes secured $3.1M from GSV and Box Group to replace outdated coding interviews with cheat-proof, project-based assessments that simulate real on-the-job tasks. Built for both staffing firms and in-house teams, Ropes uses AI to evaluate how candidates solve problems—not just whether they get the right answer.
Brisk Raises $15M to Bring AI Tools Into the Teacher Toolkit: Brisk Teaching secured $15M led by Bessemer to scale its AI-native classroom tools, following 40x revenue growth and rapid adoption across 2,000+ schools. Its Chrome extension offers AI writing detection, automated planning, and editable feedback on student work.
SchoolAI Raises $25M to Scale Personalized AI Tools for Classrooms: SchoolAI secured $25M from Insight Partners to expand its AI-powered Classroom Experience platform, which provides teachers with prep assistants and students with adaptive AI tutors. Already in over 1M classrooms, the company aims to deepen its footprint across districts while extending tools to parents and admins later this year.
Adaptive Raises $43M from a16z and OpenAI to Fight AI-Powered Threats: The platform simulates attacks, delivers real-time training, and scores organizational risk using generative AI.
Actively AI raises $22.5M to offer sales ‘superintelligence’: The startup, which just raised $22.5M from Bain and First Round, builds custom models that prioritize high-value leads by mimicking top-performing reps. Backed by results with clients like Ramp, Actively dubs its approach “GTM Superintelligence.”
Runway raises $308M Series D: to accelerate its AI video ambitions, including Gen-4, a new model for consistent, multi-perspective scene generation. Backed by General Atlantic, Nvidia, and SoftBank, the company aims to build a new “world simulation” media ecosystem.
Isomorphic Labs Raises $600M to Accelerate AI Drug Discovery: Alphabet-backed Isomorphic Labs raised $600M in its first external round to scale its AI-driven drug design platform. Known for developing AlphaFold, the company is now advancing its own clinical programs with partners like Eli Lilly and Novartis.
Tech 💻
Anthropic Opens a Window Into How LLMs “Think”: Anthropic has developed a technique to trace how its Claude model internally solves problems — revealing that LLMs don’t always reason the way we assume. Using a new technique called circuit tracing, Anthropic discovered that Claude often solves tasks in unexpected ways — like pre-planning rhymes or making up math shortcuts.
DeepMind’s Dreamer AI Finds Minecraft Diamonds (Without Help): Dreamer learned to collect diamonds in Minecraft without human help, using a world model to “imagine” outcomes. By building a world model and predicting future actions, Dreamer marks a leap toward general-purpose AI that can adapt and explore complex environments on its own.
OpenAI adopts rival Anthropic’s standard for connecting AI models to data: OpenAI is integrating Anthropic’s Model Context Protocol (MCP), enabling AI agents to pull data from apps and tools more effectively. It's a notable moment of alignment between two major foundation model players.
AI Experts Say We’re on the Wrong Path to AGI: Hundreds of top AI researchers expressed skepticism about current AGI development approaches. 76% believe simply scaling up today’s models won’t get us to human-like intelligence.
Alibaba’s Tsai Warns of AI Data Center Bubble: Tsai cautions that data center construction may be outpacing actual demand for AI services, calling out U.S. tech giants for “investing ahead of demand.”
Safety and Regulation ⚖️
“AI 2027” Warns of Rogue Agents and Global Tensions: A new report from the AI Futures Project lays out a detailed (and extreme) scenario of AI progress through 2027 — predicting superhuman coders, runaway AI researchers, and geopolitical chaos. Led by ex-OpenAI researcher Daniel Kokotajlo, the team sees autonomous, self-improving AI arriving within two to three years. While critics call it sci-fi alarmism, the project builds on eerily accurate past forecasts.
Pew: Experts and the Public See AI’s Future Very Differently: A new Pew study shows a massive perception gap—56% of experts see AI as a net positive for the U.S., versus just 17% of the public. While experts expect modest job impact, most Americans fear job losses.
Report Alleges GPT-4o Was Trained on Paywalled O’Reilly Content: While not definitive, the study’s evidence points to GPT-4o having far greater “recognition” of restricted content than earlier models like GPT-3.5.
YouTube Cuts Off Ad Revenue for AI-Generated Fake Trailers: YouTube demonetized two major channels — Screen Culture and KH Studio — for using AI to create misleading fake movie trailers that often outranked official ones.
DDoS Attacks Surge as AI Turns Them Into Geopolitical Weapons: DDoS attacks hit 17M in 2024, increasingly used in global conflicts.
Other
UFC signs wide-ranging sponsorship deal with Meta: A multimillion-dollar, multi-year partnership spanning Meta’s full product suite, including Meta AI, Quest, and Threads.
Tinder Unveils Playful AI Chatbots in Bid to Boost Engagement: Designed to boost engagement and appeal to Gen Z, the limited-time game rewards warmth and curiosity, offering real-time feedback — but stops short of full-on AI companions.
Trump’s new tariff math looks a lot like ChatGPT’s: ChatGPT, Gemini, Grok, and Claude all recommend the same tariff calculation.
A Debate Around AI, Art, and Ownership: Siegler dives into the viral Studio Ghibli-style image craze sparked by ChatGPT, using it as a lens to explore deeper questions about AI-generated art, creator intent, and what we value as "real" creativity.