DeepSeek-V3.2 🤖, Runway Gen-4.5 📹, Gemini Projects 💼

DeepSeek's V3.2 matches GPT-5, and its higher-compute V3.2-Speciale variant rivals Gemini-3.0-Pro and earned gold medals at IMO, IOI, and ICPC 2025. ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With TLDR AI 2025-12-02 Airia: Enterprise AI Orchestration — Agents, Integrations, Workflows, and Governance (Sponsor) You want AI to become part of your organizational DNA - and that means enabling every department to build out their own use cases, without IT gatekeepers standing in the way. But it shouldn't mean an ungoverned free-for-all. Airia is the "let's get serious about AI adoption" platform. Rapidly prototype, deploy, and manage AI agents that transform workflows across your organization - without sacrificing security or governance. Connect to dozens of enterprise applications with native integrations. Build agents quickly with templates and no-code tools. Let everyone build with AI while maintaining visibility and control. Plans start at just $49/month. Get a demo 🚀 Headlines & Launches DeepSeek-V3.2 (5 minute read) DeepSeek's V3.2 matches GPT-5, and its higher-compute V3.2-Speciale variant rivals Gemini-3.0-Pro and earned gold medals at IMO, IOI, and ICPC 2025. Introducing Runway Gen-4.5: A New Frontier for Video Generation (3 minute read) Gen-4.5 topped the Artificial Analysis text-to-video benchmark, ahead of Veo 3 and Sora. The model emphasizes physical accuracy (realistic momentum, fluid motion, and material coherence). Runway has acknowledged persisting issues with object permanence. Early look: Gemini's ChatGPT-style 'projects' are taking shape (3 minute read) Google has been working on a new 'projects' option for Gemini. The feature will allow users to sort files and chats around particular topics. Projects can be pinned for easy access. There is a 10-file limit, but paying users may receive a higher ceiling. 🧠 Deep Dives & Analysis A Practical Approach to Verifying Code at Scale (7 minute read) OpenAI trained an agentic code reviewer for Codex since noisy safety tools are inevitably bypassed. The system now handles 100k+ external PRs daily, providing repo-wide context and execution access. Internally, it has caught launch-blocking bugs and protected high-stakes experiments. The End of the Train-Test Split (13 minute read) Models don't know what is 'out of distribution' for themselves. Engineers will always be required in the loop until this problem is solved. It will never be easy to check the accuracy of models. Researchers need to look at their data, make sure it's clean, and label it correctly. How Can Interpretability Researchers Help AGI Go Well? (32 minute read) The Google DeepMind mechanistic interpretability team has pivoted to a pragmatic approach to interpretability over the last year. It is important to have empirical feedback on goals with good proxy tasks. Near-complete understanding isn't required for significant impact. Good focused projects start with a theory of change, and good exploratory projects start with a robustly useful setting. 🧑‍💻 Engineering & Research Cut the cost of IT complexity (Sponsor) IT complexity can drive costs up across your company. Intelligent IT automation is key to taming technology chaos. Get the insights you need from the IBM Institute for Business Value. 👉 Read the IBV report Nvidia Launches Vision Language Model for Autonomous Vehicles (4 minute read) Nvidia introduced Alpamayo-R1, the first open-source vision language action model designed for autonomous driving, at NeurIPS. Alpamayo-R1 integrates visual and textual reasoning to enhance decision-making in real-world environments. STARFlow: Scalable Transformer Auto-Regressive Flow (Hugging Face Repo) STARFlow and STARFlow-V are state-of-the-art transformer autoregressive flow models for high-quality image and video generation. STARFlow introduces a novel transformer autoregressive flow architecture that combines the expressiveness of autoregressive models with the efficiency of normalizing flows. STARFlow-V is an end-to-end video generative model with normalizing flows. Examples of generated videos and comparisons are available. Bridge Models for Image and Video Translation (GitHub Repo) ViBT introduces Vision Bridge Transformers, scaling Brownian Bridge Models to 20B parameters for efficient conditional generation. The models use a Transformer architecture and a variance-stabilized objective for robust performance on image and video editing tasks. 🎁 Miscellaneous OpenAI Takes Stake in Thrive Holdings, a Buyer of Services Firms (6 minute read) OpenAI has taken an ownership stake in Thrive Holdings to embed AI into high-volume business processes, beginning with accounting and IT services. OpenAI teams will work directly inside Thrive's companies to improve speed, accuracy, and cost efficiency. ByteDance's TikTok Playbook Is Winning Consumer AI (5 minute read) ByteDance's Doubao app is now China's most popular mobile AI platform. It received more than 11.4 million downloads in October. Doubao's focus on frictionless AI-powered voice, image, and video experiences sets it apart from competitors. ByteDance keeps its most advanced technology proprietary, breaking away from China's open-source approach. This could give it a durable commercial edge. ⚡ Quick Links Billions spent, yet AI pilots stall. (Sponsor) Discover why observability is the missing link for scaling trustworthy AI—and how it tackles hallucinations, compliance, and cost. View the findings a16z's gigawatt-scale data center timeline (1 minute read) Gigawatt-scale data centers can likely be built in 2 years or less Accenture and OpenAI Partnership (3 minute read) Accenture will deploy ChatGPT Enterprise to tens of thousands of its professionals, marking the largest upskilling effort yet via OpenAI Certifications. Black Forest Labs raises $300M at $3.25B valuation (1 minute read) The Flux image model maker, whose co-founders also created Stable Diffusion, raised funding from Salesforce Ventures, a16z, Nvidia, and others. Apple AI chief steps down following Siri setbacks (3 minute read) John Giannandrea is out after Tim Cook reportedly "lost confidence" in his leadership after repeated delays on an updated Siri. Claude Code over Excel (1 minute read) The LlamaSheets API lets users automatically segment and structure complex Excel sheets into well-formatted 2D tables. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/0b6a6dc1/2 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of AI professionals and decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! If you have any comments or feedback, just respond to this email! Thanks for reading, Andrew Tan, Ali Aminian, & Jacob Turner Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR AI isn't for you, please unsubscribe.