• Synthetic
  • Posts
  • Is New Chinese AI Even Better Than DeepSeek? 🧠

Is New Chinese AI Even Better Than DeepSeek? 🧠

Plus, The Nuclear-Level Risk of Superintelligent AI đŸ’„

Author

Steve Brown
March 13, 2025

Subscribe to Synthetic

The week’s most interesting and relevant AI news and analysis

This Week in AI

Manus, from a Chinese company that partners with Alibaba, is an autonomous, multipurpose AI agent. More accurately, Manus is an interface that controls multiple models to execute complex tasks on your behalf
.autonomously. In a promotional video, Manus is touted as ‘the world’s first general AI agent’ and is shown performing tasks including resume screening, complex property research, stock analysis, and interactive website creation. Manus can create formatted reports, spreadsheets, data visualizations, and interactive websites. It remembers user preferences, so you don’t have to keep telling it precisely how you want it to operate when you use it for everyday tasks. Reviewers at MIT Technology Review wrote that Manus occasionally crashes, and their servers are sometimes overloaded due to high demand. However, the product was highly intuitive and ‘shows real promise for the future of AI helpers.’ Manus is built on Anthropic’s Claude 3.5 Sonnet model and fine-tuned versions of Alibaba’s Qwen models.

Synthetic’s Take: Early benchmarks released by Manus claim it outperforms OpenAI’s latest Deep Research agent. Those with early access have praised Manus’ capabilities online. Still, very few people have been able to try its capabilities and put it to the test as it doesn’t yet have enough servers available to support many users. Time will tell whether Manus turns out to be the real deal or more Chinese hype.

GAIA Benchmarks Show Manus Performance (Source: Manus)

Microsoft CEO Satya Nadella distinguishes between ‘knowledge work’ and ‘knowledge workers’ as he tries to parse his language about how AI will reshape work and entire industries. He expects AI to redefine cognitive labor but not make human workers irrelevant. Instead, it will change how they work. While fears of AI job losses are widespread, early data shows that AI is a job shifter, not a job killer. The U.S. Bureau of Labor Statistics projects that professional, scientific, and technical service jobs will grow 10.5% from 2023 to 2033.

Quick Hits

  • Bank of America Estimates 3 Billion Humanoid Robots by 2060 - Predictions about the size of future humanoid robot populations vary wildly. Brett Adcock and Vinod Khosla have both predicted a market for 10 billion humanoids by 2040. Synthetic believes the number should be closer to 1 billion. Bank of America has weighed in with their prediction: 3 billion humanoids by 2060. Whichever prediction is correct, that’s still a lot of robots that will be sharing the planet with us.

  • Google DeepMind Gets Into Robots - Google released two new models in its Gemini family, Gemini Robotics and Gemini Robotics-ER (extended reasoning), that can control sophisticated robots. DeepMind has been conducting robotics research for many years, and now Google is partnering with Texas-based Apptronik to build humanoid robots using Gemini 2.0 technology. Google DeepMind is also working with Agility Robotics, Boston Dynamics, and other vendors to guide the future of Gemini Robotics-ER.

  • Robots Will Compete With Humans in Half-Marathon—Beijing will host the first half-marathon in which humanoid robots compete against people. No wheels are allowed, though remote-controlled robots are. đŸ€–

  • Feds Drop Bid To Make Google Sell AI Investments - Google won’t be forced to sell off its 14% stake in competitor Anthropic as part of an antitrust case, but it may be required to spin out its Chrome browser. 💰

Video: Mo Gawdat: AI Knows What’s Wrong With Humanity

Mo Gawdat is an entrepreneur and author and was the Chief Business Officer at Google X. He’s a widely respected AI commentator and future thinker. In this interview, he argues that AI presents humanity with an opportunity to create a world of abundance and solve many human challenges, but only if we guide and shape it with human ethics and values. This might be the most thought-provoking video you’ll watch all week. (22 minutes)

AI Tech and Innovation

Neurons, grown onto a silicon chip inside the new CL1 biological computer

Cortical Labs has launched the CL1, a computer with integrated living neurons. The company says the CL1 delivers real-time computation designed to solve more profound problems using less energy than conventional computers. Priced at $35,000, the CL1 connects real neurons to silicon chips. If you want to try out their technology, you can access it through their Cortical cloud, which offers neural computing on demand. The company plans to release a biological network server stack by the end of 2025, making the technology more widely available.

Most large language models are so big they require clusters or, often, a data center full of GPUs to run them. Google claims its latest open-source model, Gemma 3, is the best in the world to run on a single GPU or TPU. It has 98% of the accuracy of DeepSeek while running on a tiny footprint. So it can run locally on a laptop or even a phone. Gemma 3 comes in several sizes, with 1 billion, 4 billion, 12 billion, or 27 billion parameter options. Compare that to Llama 3’s 405 billion or DeepSeek’s 671 billion parameters. Google achieved this feat using a technique known as distillation. Most Gemma 3 models have a 128,000-token context window, support 140 languages, and are multimodal. To fit inside just 1GB of memory, its smallest version (Gemma 3 1B) is limited to text and English only. You can find more information in the Gemma 3 developer guide.

Gemma 3 compares favorably with much larger models

AI Insights

Special occasion dress manufacturer and wholesaler Amarra uses AI for inventory management and content creation. They make dresses and gowns for weddings, quinceañerasm, and other fancy events. They use AI to understand customer needs, write product descriptions, analyze market trends, optimize customer service, and manage inventory levels. The company was founded in 2018 and has just 25 employees. AI chatbots handle 70% of customer inquiries. Using ChatGPT to help them write product descriptions cut content creation time by 60% with no loss in quality. An AI-powered inventory system helps them set stock levels in 800 resellers worldwide. It has reduced overstocking by 40% and ensured resellers don’t sell out of popular items, both major retail issues.

Time Magazine explores what it calls ‘perhaps the most geopolitically precarious technology since the atom bomb’ and describes how the development of superintelligent AI could lead to a dangerous game of technological one-upmanship between nations. Taking inspiration from the relative peace brought about by a world of mutually assured destruction in the nuclear age, the authors propose a framework of “Mutual Assured AI Malfunction” (MAIM) as a deterrent against the risks posed by superintelligent AI. In related news, former Google CEO Eric Schmidt has sounded the alarm over a ‘Manhattan Project’ for superintelligent AI. He proposes a strategy of deterrence, transparency, and international cooperation to ensure AI benefits humanity and doesn’t cause harm.

“Agentic” is the AI buzzword of 2025, and industry watchers expect the market to be flooded with new AI agent offerings this year. Agentic AI represents a new phase of artificial intelligence that operates autonomously, making decisions and driving outcomes using tools without human intervention. The ARC framework (Accelerate, Replace, Create) offers a roadmap for AI adoption, guiding businesses from basic automation to the advanced, collaborative systems of agentic AI. The key to this process is trust. Businesses must gradually build trust with employees and customers as they deploy, reinforcing and validating that trust at every stage of deployment.

Toolkit for the Future

Generate ad banners, text, images, and video that outperform those of your competitors with better conversion and click-through rates. Get AI-powered creative insights and analyze your competitors’ campaigns to stay ahead of the curve. Free to try.

Here’s Why Over 4 Million Professionals Read Morning Brew

  • Business news explained in plain English

  • Straight facts, zero fluff, & plenty of puns

  • 100% free

Rapidly build and deploy specialized AI workers with an intelligence infrastructure that gives everyone—from developers to non-technical teams—the power to build with AI. MindStudio already powers over 100,000 AI workers at companies including Adobe, Google, IBM, and Meta. The easy-to-use, no-code, graphic workflow makes it easy for anyone to build an AI coworker. A free level is available.

This all-in-one AI-powered tool helps you stay on top of your spending and investments and offers personalized guidance and insights on managing your credit and personal finances. There’s a conversational chat interface, and new features are added each month.

Turn customer feedback into action with powerful AI-powered surveying and analytics tools. These tools allow you to measure experience across the customer journey, capture key metrics, convert feedback into growth opportunities, enhance customer satisfaction, and turn customers into loyal advocates. Trusted by brands including American Express, SAP, and Accor, and free to try.

Help your sales team stay focused on customers during meetings. Laxis captures attendees’ comments verbatim and flags items for follow-up. It auto-generates meeting summaries and follow-up emails in seconds, quickly identifying customer requirements, pain points, and action items. Free to try.

Extract data from websites and place it in a spreadsheet that fills itself and notifies you of any changes. Pre-built robots make it a breeze to get started. Extract whatever you need: job listings, property details, TikTok videos, hotel info, or competitor pricing. It's free to try, and no coding is required.