No Shortage of Experiments in AI
OpenAI’s numbers are up, Google’s making changes under the hood, and TikTok is giving photos a life of their own. Tencent, Airbnb, and Harvey are trying out fresh uses for AI, while experts admit progress may slow in some areas. Here are the facts from this week’s round of moves.
👋 Catch up on the latest post …
In the Spotlight 🔦
TikTok Unveils AI Alive, Its New Image-to-Video Creation Tool
🖼️ TikTok AI Alive enables users to transform static photos into short-form videos with AI-generated movement and creative effects within TikTok Stories.
🔖 All AI Alive videos are marked with an AI-generated label and include C2PA metadata to ensure transparency and traceability, even if shared outside TikTok.
🛡️ TikTok implements multiple safety checks and moderation for AI-generated content, allowing users to report violations and ensuring compliance with platform policies.
What to Expect at Google I/O 2025: Gemini Improvements and Android 16 Updates
Business & Industry Trends
🤖 Google I/O 2025 will feature major updates to the Gemini AI model family, including a potential new Gemini Ultra and expanded subscription tiers.
📱 Android 16 will debut with improved notifications, a new Material 3 Expressive design, enhanced accessibility, and better Bluetooth device switching via Auracast.
🛠️ Google will highlight advancements in AI agents (Astra, Project Mariner), open models (Gemma), and possibly unveil new tools like video-generating features for NotebookLM.
OpenAI Gains Ground in Enterprise Market, Surpassing Key Competitors
Business & Industry Trends
📈 OpenAI's enterprise adoption surged to 32.4% of U.S. businesses by April 2025, up from 18.9% in January, outpacing competitors.
🏆 Rivals lag behind: Anthropic reached only 8% adoption, and Google AI subscriptions dropped sharply to 0.1% by April, per Ramp’s AI Index.
💰 OpenAI projects $12.7B revenue for 2025 and expects enterprise customers to drive significant growth, despite not anticipating cash-flow positivity until 2029.
Microsoft and OpenAI Reportedly Renegotiating Partnership Terms
Business & Industry Trends
🤝 Microsoft and OpenAI are in "tough negotiations" regarding their partnership and investment structure, with Microsoft being a key holdout in OpenAI's corporate restructuring.
🏢 OpenAI's restructuring will keep its nonprofit board in control, while converting its business arm into a for-profit public benefit corporation, pending partner approval.
💼 Negotiations involve equity and technology access, with Microsoft potentially trading some equity for access to OpenAI tech developed after the current 2030 agreement cutoff.
Google Launches Initiative to Support AI-Focused Startups
Business & Industry Trends
🚀 Google launched the AI Futures Fund to invest in startups using DeepMind's latest AI tools, offering support from seed to late-stage companies.
🧑💻 Startups gain early access to Google DeepMind models, expert guidance, and Google Cloud credits, with some eligible for direct investment from Google.
🌍 No fixed application window; investments are considered on a rolling basis, and fund size or check amounts are not publicly disclosed.
How to Watch Every AI and Android Reveal at Google I/O 2025
Events & Conferences
🎤 Google I/O 2025 will be held May 20-21, featuring major announcements from Sundar Pichai, Demis Hassabis, and other Google executives.
🤖 AI advancements—especially updates to the Gemini model and Project Astra—will be a central theme, highlighting AI's integration across all Google products.
📺 Livestreams and sessions will cover topics like generative AI in Search, AI agents, autonomous vehicles, and updates to Android, Chrome, and Google Cloud.
AI Reasoning Model Improvements May Soon Slow, New Analysis Finds
Research & Development
🧠 Reasoning AI models' performance gains may slow down soon, with progress expected to converge with overall AI frontier by 2026, according to Epoch AI analysis.
💻 Reinforcement learning is being prioritized with much more computing power, but there is an upper bound to how much this can improve reasoning model performance.
💸 Scaling reasoning models faces challenges beyond compute limits, including high research overhead costs and persistent issues like hallucinations and high operational expenses.
Tencent Hires Microsoft-Affiliated WizardLM AI Team with Unusual Background
Business & Industry Trends
🤝 Tencent hired the WizardLM team, a Microsoft AI research group, to join its Hunyuan AI development organization.
🚀 WizardLM has already released a Hunyuan-branded model, Hunyuan-TurboS 0416, which claims to outperform open models like Google’s Gemma 3 series.
💰 Tencent is investing 90 billion yuan ($12.49B) in AI infrastructure in 2025, attributing 8% Q1 growth to its AI initiatives.
Anthropic and Google Gain OpenAI-Backed Startup Harvey as New User
Business & Industry Trends
🤝 Harvey, a leading legal AI startup, will now use Anthropic and Google models in addition to OpenAI, expanding its foundation model partnerships.
📈 Harvey's internal BigLaw benchmark revealed that multiple models, including non-OpenAI ones, now outperform its original system on key legal tasks.
📊 Harvey will launch a public leaderboard to transparently rank AI models' legal task performance, with detailed research insights from top lawyers.
SoundCloud Reverses Changes to AI Terms of Use
Ethics & Governance
🔄 SoundCloud revised its terms after backlash over AI-related clauses that users feared enabled training AI on their audio content.
📢 CEO Eliah Seton admitted the original terms were too broad and unclear, causing confusion about SoundCloud’s intentions regarding AI use.
🛑 Updated terms now clarify SoundCloud will not use user content to train generative AI models that replicate or synthesize voices, music, or likenesses.
OpenAI Launches GPT-4.1 Models on ChatGPT
Technology & Infrastructure
🚀 OpenAI released GPT-4.1 and GPT-4.1 mini models to ChatGPT, enhancing coding and instruction-following capabilities while offering faster performance than previous models.
💡 GPT-4.1 is now available to ChatGPT Plus, Pro, and Team subscribers, with GPT-4.1 mini accessible to all users, replacing the previous GPT-4.0 mini.
🔍 OpenAI committed to greater transparency by launching a Safety Evaluations Hub and pledging more frequent publication of internal AI model safety evaluations.
This content is curated and compiled autonomously using a Python-driven AI system I wrote. While every effort is made to ensure accuracy, I welcome your feedback on any discrepancies you may notice.