- Synthetic
- Posts
- DeepSeek and Alibaba Shake the AI World💥
DeepSeek and Alibaba Shake the AI World💥
Plus, AI Fashion Coaches and AI at the Super Bowl 🏈
Subscribe to Synthetic
This week’s most interesting and relevant AI news and analysis
This Week in AI

DeepSeek R1 Takes a Chainsaw to the Western AI Industry
It’s been a turbulent week in AI, and this week’s Synthetic reflects that with special coverage of the new DeepSeek R1 model and the turmoil it’s caused in the markets and the boardrooms of AI companies worldwide. This issue is twice the usual length but worth the read from beginning to end. Be sure to read the section on Alibaba’s new Qwen2.5-VL model, which proves the Chinese wave isn’t just about DeepSeek. 🌊
DeepSeek Continues to Dominate Headlines 📰
Last week, Synthetic covered the release of the new DeepSeek R1 reasoning model. While the model performs similarly to OpenAI’s flagship o1 model, DeepSeek claims to have trained it in just six weeks, with a team of only 200 people, using a small fraction of the computing resources OpenAI uses to train its models. Nvidia’s stock dropped almost $600 billion as the markets panicked that R1 signals a dramatic softening in demand for high-end GPUs. Incidentally, Nvidia has experienced 8 of the 10 biggest single-day stock declines in history. Other tech stocks took a battering and saw sell-offs as the AI panic spread and cooler heads failed to prevail.
There was a lot of hysteria as venture capitalist Marc Andreessen declared this was “AI’s Sputnik moment,” and Peter Diamandis hyperbolically gushed, “Nvidia’s moat just turned into a puddle.” It was rumored that Meta was perhaps scrapping plans to release its upcoming Llama 4.0 model as engineers frantically met in war rooms to determine how DeepSeek achieved these results. CEO Mark Zuckerberg says he’s not worried about DeepSeek and continues to think “that investing very heavily in CapEx and infrastructure is going to be a strategic advantage over time.” He’s probably right about that.
Here’s Synthetic’s take:
Constraint fuels innovation.
Necessity is the mother of invention. Cold War Russian programmers learned to write super-efficient algorithms because they only had access to subpar hardware. The same has happened here. The Chinese researchers at DeepSeek were forced to bypass Nvidia’s CUDA software abstraction layer and program down to the metal (programming lower-level elements in the chip) to access specific functions that enabled them to get great results out of Nvidia’s H800 chips (hobbled Nvidia H100 chips, repackaged for the Chinese market to bypass export restrictions.)
DeepSeek R1’s model is good for the whole AI industry.
AI researchers everywhere are analyzing what DeepSeek did. Their next-generation models will be far better for it. Roadmaps will accelerate, and release timelines will be shortened. Competition is always fuel for innovation. OpenAI CEO Sam Altman tweeted, “It’s legit invigorating to have a new competitor! We will pull up some releases.” The flip side of this acceleration…..safety concerns.
By lowering the barriers to entry, more bright young innovators will participate and move the industry forward at an even faster pace. A new set of AI startups will flourish, safe in the knowledge that they don’t need hundreds of millions of dollars to achieve breakthroughs.
DeepSeek R1 is actually great for Nvidia.
Scaling isn’t dead. AI progress is the result of combining scaling (e.g., data, computing, and model size) with innovation (e.g., new algorithms, architectures, and approaches). Scaling still matters, and powerful AI chips will still matter.
Jevons's Paradox states that as a product or commodity becomes cheaper, the overall money it generates doesn’t reduce; it increases, often dramatically. Fuel efficiency gains in cars increase fuel use since people drive more. If models become more affordable to train and use (due to adopting R1’s approach), people will build more models and use them far more, and demand for AI hardware will go up, not down.

Jevons Paradox illustrated (with an incorrect apostrophe!) Credit: Sketchplantations
DeepSeek R1 works quite well but isn’t a replacement for ChatGPT, Gemini, or Claude.
DeepSeek lacks the infrastructure to scale availability and meet serious demand.
DeepSeek R1 seems to send data back to China and spy on its users. It is being considered a national security risk.
The model adheres to strict Chinese ideology. Try asking it about Tiananmen Square, Uyghur oppression, or President Xi and Winnie the Pooh. 🍯
Open-source AI has caught up with closed-source AIs.
Two years ago, closed-source proprietary frontier models like ChatGPT-4, Gemini, and Claude had a comfortable 1- to 1.5-year lead over the nearest open-source models. That gap narrowed with Meta’s release of Llama 3. Now, R1 has all but erased the gap and shown that open-source efforts that build on top of previous open-source efforts can now compete at the forefront of AI development, especially if you steal work from others. 😀 OpenAI seems concerned that DeepSeek copied their work and harvested their data using a technique known as distillation.
This might accelerate the arrival of AGI. 🧠
It’s still unclear what it will take to build artificial general intelligence (human-level intelligence). It will likely require lots of powerful, expensive, energy-hungry computing, perhaps more than will be available for some time (See last week’s story on Stargate). The innovations in R1 may reduce the computing power needed to build AGI and, in a compute-constrained environment, might lead to it coming sooner. 🔜
For more on the DeepSeek news, what it means for the future of AI, and why the reaction is overblown, please check out Synthetic’s recent post, Necessity Is The Mother Of Invention.
And in the spirit of balance, here are some other opinions on what this means for the industry, and for us all:
This was a good week for DeepSeek memes. Here are some of our favorites:
![]() ![]() | ![]() ![]() ![]() |
In Other News
This is the third time that OpenAI has waited until right after the weekly release of Synthetic to break news. Last week, about an hour after Synthetic hit your inbox, OpenAI launched Operator, their first foray into the world of agents. In a normal week, this would have been the biggest news of the week, but DeepSeek had other ideas. Operator’s ability is still somewhat limited, but it can perform simple tasks using your web browser, and early reviews are optimistic about its future potential. For example, it can help you book travel, potentially disrupting the travel industry.
Synthetic’s Take: This is the year AI agents grab everybody’s attention and start to find their way into the workplace. Operator is just an opening salvo, and we expect new agents to come thick and fast through 2025, each with new and improved capabilities. Your next coworker may be a machine, and digital employees will quickly become part of every workforce.
DeepSeek isn’t all China has to offer to shake up the AI world this week. 💥 Alibaba just dropped a new model, Qwen2.5-VL, similar to OpenAI’s Operator. And it’s very impressive by all accounts. Yahoo News reports that it can analyze charts and graphics, count objects, control a PC, understand multi-hour videos, extract data from invoices, and parse files. The model also generates images that are as good as anything OpenAI’s DALL-E 3 can create, writes code, searches the web, and creates videos, too. Oh, and it’s also open-source. Chinese researchers have thrown down the gauntlet with DeepSeek R1 and now Qwen2.5-VL (though they could still use some help on a catchy name).
Synthetic’s Take: Expect Google and Anthropic to release agents in the coming weeks and frontier research labs to accelerate the release of new models to respond. We may also see a rethink of the open vs closed-source debate. Open-source advocates (like Meta and AI godfather Yann LeCun) believe open-source is the best way to build and deploy powerful models at scale (the same way open-source Linux has pervaded internet infrastructure). Closed-source advocates rightly point to safety concerns and how open-source enables geopolitical competitors.
OpenAI safety researcher Steven Adler has left OpenAI, claiming that the frontier AI lab is taking a “very risky gamble” with humanity amid the race toward artificial general intelligence (AGI). He also criticized how the race for AGI is shaping up between global superpowers. Adler shared, “Today, it seems like we’re stuck in a really bad equilibrium. Even if a lab truly wants to develop AGI responsibly, others can still cut corners to catch up, maybe disastrously. And this pushes (everyone) to speed up.”
Video: China’s DeepSeek Sparks Global AI Race
ColdFusion is a YouTube Channel known for releasing thoughtful, well-produced videos on tech and society. This episode is an excellent primer for anyone who hasn’t been keeping up with the DeepSeek moment or wants to understand the implications for the future. (18m 41s)
AI Tech and Innovation

MIT Researchers Found That Video Games With Noisy Environments
Trained Better AI Agents
When we hire new employees, we want them to be adaptable, so they can navigate unfamiliar situations and use their experience to overcome novel challenges. The same is true of AI-powered robots and agents. Researchers at MIT and elsewhere have found that training agents in different environments results in more robust agents that can navigate unfamiliar situations. MIT scientists trained agents on Atari games and found that they worked better if trained on ‘noisy’ environments. This is similar to work done by DeepMind and OpenAI some years ago.
AI Insights

There is a range of AI-powered tools to help people navigate the fast-fashion landscape and generate a specific look. These AI models understand style and complementary color palettes and can make insightful recommendations on how to dress for success. The article considers tools, including Style DNA, Aiuta, and good old ChatGPT.
AI has become an essential workplace companion for many, though integrating it into daily workflows is not as easy as we would like. A recent survey found that 39% of workers use AI to generate ideas, 37% to create content, 33% to summarize communications, and 31% to find documentation. Most workers (63%) expect AI to lead to a future with more fulfilling work and improved work-life balance. And they are already seeing the benefits of AI today: 62% saw productivity gains, 40% saw cost savings, and 38% saw improved communications and decision-making. AI must be deployed strategically to enhance work, elevate performance, and boost job satisfaction. Fundamentally, AI must empower people, not replace them.
The U.S. copyright office has declared that using AI-powered tools in the creative process does NOT undermine the copyright of a work. This ruling, delivered in this 41-page report, clears the way for continued (read, expanded) adoption of AI-powered post-production tools in filmmaking. Last week, Synthetic reported the controversy around AI’s use to enhance scenes in Oscar-nominated movies The Brutalist and Emilia Perez.
Get ready for AI to be everywhere in the commercial breaks of the Super Bowl this year. Expect AI-focused spots from tech giants like Microsoft and Google and lots of AI-generated creative from other brands. With Super Bowl spots costing up to $8 million per 30 seconds, let’s hope it’s money well spent. 🏈🍿
There’s a reason 400,000 professionals read this daily.
Join The AI Report, trusted by 400,000+ professionals at Google, Microsoft, and OpenAI. Get daily insights, tools, and strategies to master practical AI skills that drive results.
Toolkit for the Future
Create and sell online courses from your own website with this top-rated platform. Easy to use with hundreds of beautiful templates and a powerful generative AI assistant that makes content generation a breeze. The AI will create lessons, exams, assessments, and more. Free to try.
Relieve pressure on your support team and serve customers better and faster with a support automation platform. AI answers 90% of queries and escalates the rest to your team. It automatically updates its knowledge database so it’ll never escalate that same issue again. Supports SMS, social, email, and phone communication.
Focus on the email that matters. Deep Clean automatically identifies old, unimportant emails so you can delete them. Snooze emails until a better time and easily track emails that need a response. Automatically filter unimportant emails into specific folders. It works across every device where you check your email. Get to inbox zero with award-winning SaneBox.
Need fresh talent that can harness the latest AI tools to increase the speed and volume of marketing content creation? Unlock value with fully-vetted, top marketing people with experience in your industry. The marketing-as-a-service model allows you to hire talent on demand, part -time or full-time, in 30-day blocks of time.
Try this easy-to-use, AI-powered alternative to DocuSign to write proposals, create quotes, get eSignatures on contracts, build online forms, collect payments, and much more. Great for sales, HR, legal, finance, and IT. Free 14-day trial.