• NATURAL 20
  • Posts
  • China’s DeepSeek Plans AI Agent to Rival OpenAI by 2025

China’s DeepSeek Plans AI Agent to Rival OpenAI by 2025

PLUS: Google Photos Adds Veo 3 — Turn Pics Into 4-Second Videos, Sierra Raises $350M, Hits $10B Valuation in Just 18 Months and more.

In partnership with

Kickstart your holiday campaigns

CTV should be central to any growth marketer’s Q4 strategy. And with Roku Ads Manager, launching high-performing holiday campaigns is simple and effective.

With our intuitive interface, you can set up A/B tests to dial in the most effective messages and offers, then drive direct on-screen purchases via the remote with shoppable Action Ads that integrate with your Shopify store for a seamless checkout experience.

Don’t wait to get started. Streaming on Roku picks up sharply in early October. By launching your campaign now, you can capture early shopping demand and be top of mind as the seasonal spirit kicks in.

Get a $500 ad credit when you spend your first $500 today with code: ROKUADS500. Terms apply.

Today:

  • China’s DeepSeek Plans AI Agent to Rival OpenAI by 2025

  • OpenAI Launches Jobs Platform & Free Training to Certify 10M Americans

  • Anthropic Bans AI Sales to Majority Chinese-Owned Firms

  • Google Photos Adds Veo 3 — Turn Pics Into 4-Second Videos

  • Sierra Raises $350M, Hits $10B Valuation in Just 18 Months

Chinese startup DeepSeek, based in Hangzhou, plans to release by late 2025 an artificial intelligence “agent” — a program that can act on behalf of a user. The system will handle many tasks in sequence with only brief instructions and will refine performance by studying past actions. DeepSeek aims to match or beat US leaders like OpenAI and carve out a place in the next global wave of AI tools.

Why this matters

  1. Fresh competition – A well-funded Chinese player widens the field beyond US titans, spurring faster innovation.

  2. Shift toward agents – Emphasis on multi-step, self-improving helpers highlights the next big phase in how people will use AI.

  3. Aggressive timeline – Targeting end-2025 sets a clear marker for investors, policymakers, and researchers tracking AI’s rapid progress.

OpenAI’s new plan aims to give more people work options as AI spreads. It will launch a Jobs Platform that matches firms with workers trained in ChatGPT and related tools. To prove skills, OpenAI will offer free online lessons and official certificates—formal proof of ability—through its Academy, pledging to certify 10 million Americans by 2030. Partnerships with Walmart, Deere, consulting firms, and state groups will help turn training into real paychecks.

Why this matters

  1. Bridges skill gap – Gives workers clear, affordable paths to learn AI basics and show verified skill, easing talent shortages.

  2. Creates hiring standard – Certificates set a shared yardstick for “AI fluency,” guiding employers and educators worldwide.

  3. Signals broad adoption – Big-name partners and a 10-million target show AI’s move from niche tool to everyday workplace necessity.

Anthropic, maker of the Claude chatbot, will no longer sell its AI services to companies that are majority-owned by Chinese firms, or to groups in Russia, Iran, or North Korea. The rule aims to block those governments from using cutting-edge AI for military or spy work. The policy, effective now, covers direct customers and cloud-based users of Claude. Anthropic expects a small revenue loss but hopes to set a standard.

Why this matters

  1. National-security spotlight – Shows labs may self-restrict sales when they fear state misuse of advanced models.

  2. Industry precedent – Could push rivals like OpenAI and Google to adopt similar country-based limits, shaping global AI trade.

  3. Tech-policy tension – Highlights growing clash between open commercial markets and government concerns over AI’s military value.

🧠RESEARCH

UI-TARS-2 is a powerful GUI-focused AI agent that uses multi-turn reinforcement learning and scalable training methods to interact with digital environments like file systems and web interfaces. It outperforms top models on key benchmarks, approaches human-level game performance, and shows strong generalization across real-world tasks and software workflows.

SimpleTIR boosts tool-integrated reasoning in large language models by solving a major training problem—unstable multi-turn reinforcement learning. It filters out unproductive “void turns” that cause training failures, leading to more stable learning. The result: record-breaking math benchmark scores and smarter reasoning strategies like self-correction and cross-validation without supervision.

LLaVA-Critic-R1 turns a model meant for judging AI outputs into a powerful generator through reinforcement learning on preference data. It outperforms many vision-language models in both reasoning and generation, proving critics can be strong policy models. Self-critique at test time further boosts performance without extra training.

🛠️TOP TOOLS

Sendsteps AI - AI-powered presentation maker that streamlines the creation of engaging, interactive presentations in minutes.

Dumme - AI-powered platform designed to automatically transform long-form videos and podcasts into short-form clips suitable for social media platforms.

BigJPG - AI-powered image upscaling tool that leverages Deep Convolutional Neural Networks to enlarge images while maintaining quality.

AICheatCheck - Educational tool designed to detect AI-generated content in academic settings.

PixaMotion - Mobile application designed to breathe life into static images through animation and motion effects.

📲SOCIAL MEDIA

🗞️MORE NEWS

  • Google Photos now uses Veo 3 to turn still pictures into silent four-second videos. Free users tap “Subtle movements” or “I’m feeling lucky.” The update boosts detail clarity, formerly called fidelity, and needs no subscription.

  • Sierra, an AI startup that makes chat-based customer-service helpers for big companies, raised $350 million, putting the firm’s worth at $10 billion. Founders Bret Taylor and Clay Bavor say the 18-month-old company already serves hundreds of clients.

  • Warner Bros. Discovery sued Midjourney, saying the AI tool creates copies and alterations of characters such as Batman and Bugs Bunny, violating copyright (ownership rights). It seeks damages and a court ban on unauthorized use.

  • Google’s Nano Banana, a photo-editing feature in the Gemini app, powered 200 million edits and attracted 10 million first-time users within a week, pushing the app close to the top spots in Apple and Android download charts.

  • AI cloud-infrastructure firm Lambda, which rents GPUs on demand, has hired Morgan Stanley, J.P. Morgan, and Citi to prepare an IPO as soon as early 2026. Backed by Nvidia, it has raised $1.7 billion and trails rival CoreWeave, which went public in March.

  • Hitachi Energy will spend $1 billion on U.S. power-grid upgrades, including a Virginia factory for large transformers. The firm credits Trump’s AI Action Plan, which drives huge home-grown energy projects and manufacturing returns.

What'd you think of today's edition?

Login or Subscribe to participate in polls.

Reply

or to participate.