DeepSeek vs. OpenAI: The AI Arms Race Gets Messy: Hashtag Trending for Thursday, January 30th, 2025

Share post:

DeepSeek vs. OpenAI: The AI Arms Race Gets Messy, Even More Competition in AI, and Frustrated Website Owners Strike Back At AI Models. 

Welcome to a special AI edition of Hashtag Trending. I”m your host Jim Love, let’s get into it. 

DeepSeek vs. OpenAI: The AI Arms Race Gets Messy

OpenAI is accusing Chinese AI startup DeepSeek of using its own AI-generated outputs to train a competing model—essentially copying a system that has itself faced accusations of using vast amounts of copyrighted data.

At the heart of the dispute is model distillation, a process where one AI model learns from the outputs of another, effectively mimicking its reasoning and style. OpenAI claims DeepSeek leveraged ChatGPT-4o’s responses to build its own AI assistant, which has rapidly gained popularity, topping app store charts in China. Some users even reported that DeepSeek’s chatbot had openly stated it was trained on ChatGPT-4o, adding fuel to the allegations.

DeepSeek’s meteoric rise—reportedly developing a model for just $6 million compared to OpenAI’s billions—has rattled the industry. Nvidia’s, the company that provides the chips that drive the vast majority of AI infrastructure, had their stock experience a sharp decline, falling between 12% and 13%, erasing over $460 billion to $500 billion in market capitalization. This marks one of the largest losses in U.S. stock market history124.

  • Other Tech Giants: Companies like Meta and Alphabet also faced significant losses, contributing to a broader selloff in the sector35.
  • Overall Market Impact: The Nasdaq 100 futures fell as much as 5.2% in overnight trading, and the combined market capitalization loss for U.S. and European tech stocks is estimated at $1 trillion4

 Investors and analysts are watching closely as the U.S. weighs policy responses. The US government has been watching closely as well. The US Navy has banned its use. The White House has announced it’s looking at the national security issues. 

Some business leaders see it as a wake-up call. For Google CEO Eric Schmidt calls it a “turning point” in the AI race and is urging the US to embrace open-source AI. Schmidt is an investor in a number of AI companies, including  Augment, a rival to GitHub Copilot that uses open models.

Last year Schmidt said that he thought the US was two or three years ahead of China. In a recent interview he revised that opinion. 

Meanwhile, OpenAI’s complaint raises deeper questions: If AI companies use public data to train their models, who really owns what in this new AI ecosystem?

The irony isn’t lost on many: OpenAI, long accused of training on copyrighted content, is now trying to draw the line on AI learning from its AI.  

If OpenAI argues that training AI on AI outputs is theft, then it risks admitting that its own data practices were questionable from the start. If it allows it, it opens the door to competitors rapidly cloning its work with minimal investment.

The real issue is that AI companies want it both ways: unrestricted access to data when building their models but strict control over how others use their outputs. This legal and ethical contradiction is shaping up to be one of the biggest fights in AI.

So, yes—the pot can call the kettle black. The real question is whether the kettle has a case to push back. 

The Damn May Be Breaking On AI Competition

The recent advancements by DeepSeek have indeed spurred a wave of competition in the AI industry. 

Alibaba has introduced its latest AI model, Qwen 2.5-Max, which it claims surpasses DeepSeek’s V3 model and OpenAI’s GPT-4o in various benchmarks, including Arena-Hard, LiveBench, LiveCodeBench, and MMLU-Pro. 

Alibaba asserts that Qwen 2.5-Max is comparable to Anthropic’s flagship Claude-3.5-Sonnet and “almost comprehensively surpasses GPT-4o, DeepSeek-V3, and Llama-3.1-405B.” 

We were able to do a quick assessment prior to recording the podcast and it does appear that Alibaba’s model can handle some logic problems very effectively. It struggled with how many rs were in strawberry, but in fairness, DeepSeek had the same problem, although when pressed, DeepSeek did an analysis and found out it was wrong, while Alibaba’s Qwen insisted it’s answer was right. 

But Qwen, however effective it may be, is a closed source model. Its prime advantage is that it will integrate with a number of other tools used by Alibaba in ecommerce. 

Others are focusing on even cheaper alternatives to DeepSeek, if that could be possible.

A research team from the University of California, Berkeley, led by Ph.D. candidate Jiayi Pan, claims to have replicated the core technologies of DeepSeek’s R1-Zero model for just $30. Their relatively small R1-Zero model, with 3 billion parameters, developed self-verification and search abilities through reinforcement learning, demonstrating remarkable problem-solving capabilities. 

No doubt there will be even more offerings in the coming days and weeks. While we can’t count out the large US providers who have invested billions of dollars in their AI solutions, there is no doubt that the competition is going to increase. 

Can US providers survive a competitive attack on their expensive models? This race to the bottom in terms of cost is undoubtedly going to have an impact on what AI providers can charge. But will it be a death blow to the US AI development which has had a virtual monopoly on AI for the past few years? What will it do to those huge valuations that have made these companies worth trillions of dollars in their market cap?  Many are claiming, surprisingly, that in the long run, it won’t have an big impact, quoting something called Jevon’s , which says that as the efficiency of technology brings the price down, usage increases. 

So as the price goes down, usage increases exponentially. 

And if that weren’t enough of a headache for the big AI providers.

Websites Fight Back: AI Scrapers Get Stuck in Digital Tarpits

As AI companies scramble to train their models on massive datasets, website owners are deploying digital tarpits—traps designed to mislead and slow down unauthorized AI scrapers. These tactics don’t just block scrapers; they actively poison AI training data, feeding models misleading, junk, or outright false information.  In a recent article by Ars Technica, cybersecurity expert Dan Petro from Bishop Fox discusses the deployment of digital tarpits to combat unauthorized AI scrapers. These tarpits serve misleading or false data to scrapers that ignore robots.txt directives, effectively poisoning AI training datasets.

Tarpits have long been used in cybersecurity to combat spammers and bots, but now they’re being adapted to disrupt AI models that ignore robots.txt—the web standard that tells bots which content they can and can’t access. Some sites lure scrapers into infinite loops, while others generate fake data meant to corrupt AI datasets. The goal? Make scraping so inefficient and unreliable that AI companies either comply with content permissions or risk training their models on polluted data.

This fight is part of a broader backlash from artists, publishers, and developers who feel AI companies are exploiting their work without consent. Some have already experimented with “glitching” images to confuse AI vision models, while others introduce synthetic adversarial data to mislead AI into learning incorrect patterns. If these efforts scale up, AI firms may be forced to develop new filtering techniques—or negotiate with content creators to avoid mass data poisoning.

With AI scrapers and digital tarpits now locked in an escalating battle, the internet itself is becoming a contested space. 

In a tribute to the inventor’s of DeepSeek, we will end today’s episode with what everyone thinks of as a Chinese proverb – or is it a curse?  May you live in interesting times. The truth is, that phrase was invented by an American but years later everyone thinks its a Chinese invention. Hmmm. I hope that’s not a prediction about AI. 

That’s our show for today. Remember to check in this weekend when Project Synapse returns with a deeper dive into the DeepSeek and the AI stories from this week. 

You can reach me at editorial and technewsday.ca or leave me a comment under this video if you are watching on YouTube.

I’m your host, Jim Love. Have a Thrilling Thursday. 

SUBSCRIBE NOW

Related articles

North Korean Hackers Trick Employees With New Social Engineering

North Korean Hackers Trick Employees With New Social Engineering, New Prompt Injection Attack Compromises Gemini's Long-Term Memorym Canada's...

Canada’s Tech Sector Faces Continuing Talent Crunch: Hashtag Trending

Report Says Canada's Tech Sector Faces Continuing Talent Crunch Amid Rapid AI Advancements, Study Reveals reCAPTCHA's Lousy At...

homson Reuters Wins Landmark AI Copyright Case: Hashtag Trending for Thursday, February 13, 2025

Thomson Reuters Wins Landmark AI Copyright Case, Tumblr joins the fediverse and converts to WordPress, The US and...

Scammers Exploit DeepSeek Hype: Cyber Security Today

Scammers Exploit DeepSeek Hype with Fake Websites and Crypto Schemes, A Researcher Jailbreaks OpenAI’s o3-mini Model, Bypassing Safety...

Become a member

New, Relevant Tech Stories. Our article selection is done by industry professionals. Our writers summarize them to give you the key takeaways