Posted on 01/27/2025 3:47:17 AM PST by RoosterRedux
A Chinese AI company that rivals ChatGPT, is gaining attention in Silicon Valley with its rapid rise, nearly outperforming leading American AI companies like OpenAI and Meta.
DeepSeek is a Chinese AI startup that develops open-source large language models (LLMs), according to the company's website.
The company unveiled R1, a specialized model designed for complex problem-solving, on Jan. 20, which "zoomed to the global top 10 in performance," and was built far more rapidly, with fewer, less powerful AI chips, at a much lower cost than other U.S. models, according to the Wall Street Journal.
The announcement of the latest version of the app happened on President Donald Trump's Inauguration Day as another Chinese-owned social media app, TikTok, was making headlines about whether it would be banned in the U.S.
...
"Deepseek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen," Marc Andreessen, a Silicon Valley venture capitalist who has been advising Trump, wrote in a post on X. "And as open source, a profound gift to the world."
"DeepSeek R1 is AI's Sputnik moment," Andreessen wrote in a comment.
Alexandr Wang, CEO at Scale AI, a San Francisco-based software company, also spoke out on the technology and said DeepSeek's quick success is a "wake-up call for America."
"DeepSeek is a wake up call for America, but it doesn’t change the strategy," Wang wrote in a post on X.
(Excerpt) Read more at msn.com ...
"Chip makers and Big Tech stocks also fell early Monday after mostly starting 2025 in the green.DeepSeek said it spent just $5.6 million to train its open-source R1 model, although that excludes some crucial expenses such as research and development.
If it is indeed possible to produce large language models able to rival that of ChatGPT for billions less than is currently being spent by Silicon Valley, these AI plays may have less demand for their products from the likes of hyperscalers such as Microsoft, Amazon, and Alphabet—and their days as market darlings could be over."
Wow china must have imported a lot of indians to make their technology so cheap and effective.
Never trust the CCP.
Trust them, no, but they are cranking out engineers while we are graduating gender studies majors who have never read a book. The threat is real.
Smart software will beat hardware everyday.
Just got to Walmart and see how many products are made in CCP country
This is why Stargate is needed.
Seems like exactly the sort of thing we’d have slipped to them to justify our side of a massive AI arms race.
Time to buy tech stocks on the dip.
It’s entirely plausible that Deep Seek’s chatbot could be the product of a hybrid approach, blending solid engineering and training practices with unauthorized use of proprietary resources to cut costs. For instance, they might have obtained stolen pre-trained weights from a state-of-the-art model, bypassing the most expensive and compute-intensive stages of development. This would leave them with the comparatively affordable task of fine-tuning the model on their own data.
Alternatively, they could have used open-source models like LLaMA as a foundation, enhancing them with stolen datasets or components from proprietary systems to boost performance. They might also have engaged in reverse engineering, querying advanced public APIs to train a model that mimics the original.
By combining these strategies with cost-saving measures like using low-cost compute infrastructure or optimizing training processes, Deep Seek could have developed a highly capable chatbot for just $5.6 million. While technically challenging, this approach is entirely feasible, assuming access to skilled developers and sufficient resources.
Too bad it’s too late to take back the view that China is backward failed country; that view caused a lot of technology transfers in the name of bringing China up to modern standards.
Wow china must have imported a lot of indians to make their technology so cheap and effective.
—
they hired recent college graduates not professionals
Stargate is needed but what the Chinese did is noteworthy as they have smaller models that took advantage of less powerful chips. So they essentially had to be efficient in their LLM right from the start.
The race for dominance in AI will have only one winner. There may be no catching up for the ones who cross the finish line even one hour after the first-place team.
It’s gonna be a brave new world.
Thx. That was interesting.
Yeesh, I know. And I think we all know that President Trump fully knows about everything CCP and how we need to completely get rid of, get it and every single thing it manufactures out of our Constitutional Republic.
Dear Lord Jesus Christ, please help President Trump and his Administration know exactly what to do and when to do it, and with Your Mighty Arm sweep these infiltrators and indeed any who collaborate with them out of the United States despite any attempts to reverse our becoming a Constitutional Republic again.
Please, Lord God, protect this Country from attack from within and without. Protect our newly-elected leaders, let them follow You and daily carry Your cross, Jesus.
Let We The People do likewise and carry the cross of Christ daily in our lives, keeping the Faith and completing the race set before us in reaching others with the Gospel. We as believers know that Your Return is so very near, as indeed the world behaves as in the days of Noah before You sent the Flood. As You tell us in Matthew 24 of the signs and behaviors of mankind.
Dear Lord God, also a prayer toward the children in our public school system, that a vast change occur within it so that these babies and youngsters and young adults who are there to learn and wish to learn in a safe environment have the ability to do so without any threat of harm. From anyone. That includes any teachers, staff, and administration who seek to harm any child, let it be prevented by Your Mighty Hand.
I praise Your Holy Name, Jesus, for the ability of parents to school their children via homeschooling, such a tremendous Blessing! For Christian schools as well, let them increase and flourish in number throughout the United States.
Thank You, Lord God, for all the Blessings and Love and Protection over us, and Your Salvation available to all who call upon Your Name. For all that call upon Your Name, Jesus, and believe that God has raised You from the dead, they will be saved.
Thank You, again, Lord God.
Amen.
DeepSeek’s R1 model is highly competitive with U.S. platforms like OpenAI’s GPT-4 for specialized tasks, particularly in logical inference, mathematical reasoning, and problem-solving. However, in specific scenarios, such as solving problems from the American Invitational Mathematics Examination (AIME), R1 is reportedly slower than OpenAI’s o1 model. This speed difference likely stems from R1’s Mixture of Experts (MoE) architecture, where only a subset of parameters is activated for each task. While this design reduces costs and computational demands, it introduces latency in routing tokens to the appropriate experts.
Additionally, R1 prioritizes accuracy and logical consistency over speed, making it ideal for high-precision tasks but less optimized for real-time performance. Its reliance on cost-effective NVIDIA H800 chips, compared to the cutting-edge GPUs used by OpenAI, may further impact inference speed. Challenges like potential language-mixing in outputs and additional formatting steps could also contribute to delays.
Ultimately, while OpenAI’s models may have an edge in speed, R1 remains a strong contender for tasks requiring accuracy and reasoning. Its innovative design and resource-efficient approach make it an impressive and cost-effective alternative to dense architectures like GPT-4, especially for applications where precision is more critical than response time.
NVDA is toast if even half of the DeepSeek claims are true.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.