🛎️The King of Cheap
Microsoft‘s New FP4 Training Framework vs. DeepSeek: Which Training Cost is Lower?
Read the storyDeepSeek
Reporting and context on DeepSeek, with a closer look at the decisions and developments that matter.
Follow DeepSeek
Explore the coverage Get AI Secret’s free daily briefing, including DeepSeek and the wider AI industry.
At a glance
DeepSeek develops AI language and reasoning models. Coverage follows its research, open model releases and their implications for the AI industry.
- Focus
- Language models; reasoning; open model releases
- Reference
- DeepSeek
Sources checked . Selected company background and milestones.
Company milestones
Selected milestones, newest first. Earlier history is available below.
Show earlier historyHide earlier history · 0 milestones
More than 10 years ago.
The latest on DeepSeek
Newest firstRelated reporting
Relevant sections from the articles on this page. Explore older coverage below.
More coverage
Hello, AI Enthusiasts!
Who will emerge as the true king of cost-effectiveness? Will it be the corporate giant with a penchant for PowerPoint presentations or the scrappy upstart looking to make its mark?
SPONSORED BY
HEADLINES
Microsoft‘s New FP4 Training Framework vs. DeepSeek: Which Training Cost is Lower?
The recent launch of Microsoft's FP4 training framework has sent ripples through the artificial intelligence (AI) and machine learning (ML) landscape. Promising a significant reduction in costs associated with training large language models (LLMs), FP4 challenges existing paradigms, particularly when compared to notable alternatives like DeepSeek.
- FP4 stands for 4-bit floating-point quantization, which minimizes the data precision used in training AI models to just 4 bits, significantly enhancing computational speed and reducing costs.
- Microsoft introduces two groundbreaking innovations: the Differentiable Quantization Estimator, which minimizes quantization errors, and the Outlier Handling Mechanism, which manages data outliers effectively to maintain model performance.
- Utilizing an FP8 mixed precision framework and a DualPipe system, DeepSeek optimizes training costs and enhances computational efficiency, while employing Mixture-of-Experts techniques and reinforcement learning for performance improvements.
- The costs associated with DeepSeek's training and inference are well-documented, while Microsoft’s FP4 lacks explicit financial details. However, FP4's 4-bit precision may potentially lower expenses and operational costs.
Microsoft‘s New FP4 Training Framework vs. DeepSeek: Which Training Cost is Lower?
The recent launch of Microsoft's FP4 training framework has sent ripples through the artificial intelligence (AI) and machine learning (ML) landscape. Promising a significant reduction in costs associated with training large language models (LLMs), FP4 challenges existing paradigms, particularly when compared to notable alternatives like DeepSeek.
Microsoft‘s New FP4 Training Framework vs. DeepSeek: Which Training Cost is Lower?
Utilizing an FP8 mixed precision framework and a DualPipe system, DeepSeek optimizes training costs and enhances computational efficiency, while employing Mixture-of-Experts techniques and reinforcement learning for performance improvements.
TOGETHER WITH TRADE
The Easiest Way To Upgrade Your Coffee At Home
I don’t know about you, but like most Americans, the first thing I reach for every morning is my favorite mug and the coffee in my pantry. Until recently, I settled for any coffee that seemed “okay” at the grocery store, hoping it’d do the job of waking me up. I didn’t care about its flavor or quality. Then I discovered Trade, and everything changed.
Trade introduced me to some of the most exciting small-batch roasters and coffee shops across the country—the kinds of places I’d love to visit if I lived nearby. I took their quick quiz about my coffee preferences, and they matched me with unique coffees I’d never find at the grocery store. Now, I can’t imagine my mornings without them.
Each bag is roasted fresh and shipped directly to my door. I was amazed at how much better it tasted than my usual grocery store coffee, and I didn’t have to change my brewing routine at all. The aroma? Even my mail carrier noticed and asked about it—it was that good. I’m never going back to the grocery store for coffee again.
For a limited time, AI Secret readers can enjoy 30% off their first order when they sign up for a Trade subscription. Simply share your preferences—such as taste, grind, and coffee consumption—and find your perfect roast. Then, sit back, relax, and enjoy fresh coffee delivered right to your doorstep. There’s no risk, only the reward of exceptional coffee.
HEADLINES
U.S. Copyright Office Sets Clear Rules on AI-Generated Content
As AI-generated content becomes more prevalent, the U.S. Copyright Office is actively clarifying its position on the copyrightability of such works, balancing the need for innovation with the preservation of human creativity.
- The U.S. Copyright Office has affirmed that copyright protection hinges on human creativity, distinguishing between AI as a tool and works generated purely by machines. A notable case saw the rejection of copyright for "A Recent Entrance to Paradise," emphasizing the necessity of human involvement.
- The Office's January 2025 report highlighted that AI-assisted creations could be copyrighted if they exhibit perceptible human creativity. Merely prompting an AI does not suffice; substantial human contributions are essential for copyright eligibility.
- After reviewing extensive public input, the Copyright Office determined that current laws adequately address AI challenges, finding no immediate need for legislative updates while acknowledging the need for ongoing assessment of technological advancements.
QUICK HITS
- Anthropic CEO Dario Amodei emphasizes the need for stronger US export controls on AI chips to China amid DeepSeek's advancements in AI technology.
- DeepSeek faces a trademark conflict in the U.S. with Delaware-based Delson Group, which filed for the "DeepSeek" trademark before them.
- Unitree’s humanoid robots performed the Chinese folk dance Yangge at the 2025 Spring Festival Gala.https://aisecret.us/introducing-chatgpt-gov-the-us-governments-most-efficient-civil-servant-yet/
- Luma AI has launched Upscale to 4K, allowing users to enhance videos up to 4K resolution.
- Google's report warns that cybercriminals are increasingly using AI to enhance fraud, hacking, and disinformation tactics.
TRENDING TOOLS
- 🎥 Riverside records high-quality podcasts and videos remotely.
- 🔠 TalkPal AI is the most efficient way to learn a language.
- 📲 Type Prompt generates human-like social posts instantly.
- 🌍 Hello8 translates your videos into 29+ languages and reach the entire world.
- 🧭 Glitter AI turns any process into a step-by-step guide.
- 🎧 Tad AI creates original songs with your choice of genres and moods using text prompts. (Featured)
- 📚 Heardly is the Fast Way to read Best Book. (Featured)
- 🦾 Flot AI writes, reads, and remembers across any apps and webs. (Featured)
- 📝 CopyOwl AI writes high-quality articles with credible sources. (Featured)
1 Million+ AI enthusiasts are eager to learn about your product.
AI Secret is the world’s #1 AI Newsletter, boasting over 1 million readers from leading companies such as OpenAI, Google, Meta, and Microsoft. We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?
What We Can Offer:
- Launch an Advertising Campaign
- Introduce New Product or Features
- Other Business Cooperation
Email our co-founder Mark directly at [email protected] if the button fails.
Introducing ChatGPT Gov: The US Government’s Most Efficient Civil Servant Yet?
The future of government work might just be here, and it’s not wearing a suit or carrying a briefcase. Meet ChatGPT Gov, OpenAI’s tailored version of its flagship AI chatbot, designed specifically for U.S. government agencies. This new AI-powered civil servant promises to revolutionize public sector operations
Magna
Alibaba Claims Its New Qwen Model Outperforms DeepSeek
Comparison Of Qwen2.5-Max, DeepSeek-V3, GPT-4o, Claude-3.5-Sonnet, And Llama-3.1-405B
Magna
Unveiling the DeepSeek Creator: Hedge Fund Billionaire Liang Wenfeng
In the world of artificial intelligence (AI), where Silicon Valley giants like OpenAI and Google have long dominated the headlines, a new name has emerged from the East to challenge the status quo: Liang Wenfeng. The 40-year-old hedge fund manager who founded DeepSeek, a Chinese AI startup, has not only
Magna
Hello, AI Enthusiasts!
The U.S. government is now getting a digital sidekick to help it navigate the labyrinthine world of bureaucracy. And don't be surprised if your toaster decides to run for Congress next year!
SPONSORED BY
HEADLINES
Introducing ChatGPT Gov: The US Government’s Most Efficient Civil Servant Yet?
The emergence of ChatGPT Gov marks a pivotal moment in the evolution of government work, integrating advanced AI technology into public sector operations. This innovative chatbot, specifically designed for U.S. government agencies, aims to enhance efficiency, security, and productivity, challenging the conventional image of civil service.
- ChatGPT Gov is a tailored version of OpenAI's AI chatbot, crafted specifically for government agencies, leveraging the infrastructure of ChatGPT Enterprise.
- The chatbot aims to alleviate the burdens of bureaucratic inefficiencies through automation and streamlined workflows.
- Over 3,500 government agencies have adopted ChatGPT Gov, with 90,000 users generating 18 million prompts, indicating a significant reliance on AI in the public sector.
Introducing ChatGPT Gov: The US Government’s Most Efficient Civil Servant Yet?
The emergence of ChatGPT Gov marks a pivotal moment in the evolution of government work, integrating advanced AI technology into public sector operations. This innovative chatbot, specifically designed for U.S. government agencies, aims to enhance efficiency, security, and productivity, challenging the conventional image of civil service.
TOGETHER WITH SQUARESPACE
New Tools To Grow Your Future Online Presence
Empowering millions of individuals from local artists to entrepreneurs, Squarespace’s elegant solutions have set a new standard for online publishing.
Design a completely custom website with Squarespace's professionally curated layout and styling options designed to sell anything. Start with a flexible designer template or build your own, then customize to fit your style and needs using our drag-and-drop website builder.
Everything you need to run an ecommerce business, book appointments, or sell your skills — all on a single platform built just for you.
- Squarespace Payments - Native payment solution offering fast and secure payment options
- Courses - Turn your expertise into income by creating and selling an online course
- Client Invoicing - Manage your business on Squarespace from vetting leads to receiving payment via invoices in one streamlined, customizable workflow.
- Acuity Scheduling - Customizable appointment scheduling made easy, automating workflows, payments, and bookings.
Exclusive benefit for AI Secret readers: Use code 'SAVE10' at checkout to get 10% off on all plans
HEADLINES
Alibaba Claims Its New Qwen Model Outperforms DeepSeek
- Comparison Of Qwen2.5-Max, DeepSeek-V3, GPT-4o, Claude-3.5-Sonnet, And Llama-3.1-405B
DeepSeek's recent success with DeepSeek-V3 shook the tech world, prompting Alibaba to respond robustly to reclaim its position. The fierce competition in the artificial intelligence landscape reaches new heights with Alibaba's recent launch of Qwen2.5-Max, a model designed to surpass leading contenders like DeepSeek-V3 and GPT-4o.
- Qwen2.5-Max features a sophisticated architecture with pretraining on over 20 trillion tokens, positioning it as a formidable competitor in the AI space.
- Alibaba claims Qwen2.5-Max surpasses DeepSeek-V3 in several benchmarks, particularly in coding tasks, showcasing its effectiveness against top models like GPT-4o.
- One of the model's standout features is its ability to run locally at minimal cost, making it an appealing choice for developers and businesses.
- Each competitor has its strengths and weaknesses, with GPT-4o and Claude-3.5-Sonnet excelling in specific benchmarks but facing accessibility issues compared to Qwen2.5-Max and DeepSeek-V3.
Alibaba Claims Its New Qwen Model Outperforms DeepSeek
DeepSeek's recent success with DeepSeek-V3 shook the tech world, prompting Alibaba to respond robustly to reclaim its position. The fierce competition in the artificial intelligence landscape reaches new heights with Alibaba's recent launch of Qwen2.5-Max, a model designed to surpass leading contenders like DeepSeek-V3 and GPT-4o.
Alibaba Claims Its New Qwen Model Outperforms DeepSeek
DeepSeek's recent success with DeepSeek-V3 shook the tech world, prompting Alibaba to respond robustly to reclaim its position. The fierce competition in the artificial intelligence landscape reaches new heights with Alibaba's recent launch of Qwen2.5-Max, a model designed to surpass leading contenders like DeepSeek-V3 and GPT-4o.
QUICK HITS
- Sam Altman praises deepseek's r1 model while promising improved models from his own team and welcomes the competition.
- Abu Dhabi's M42 is committed to strengthening ties with the US.
- Elon Musk is implementing a Twitter-style approach to streamline the federal workforce under Trump's administration, urging employees to resign if they don't comply with new office policies.
- Microsoft is investigating whether a group linked to DeepSeek improperly obtained OpenAI data.
- Convergence AI has launched Proxy, a natural language agent touted as ‘Europe’s answer to OpenAI’s Operator.’
TRENDING TOOLS
- 📀 TemPolor creates music the way you want.
- 🔎 Athina AI automatically detect and fix hallucinations in prod.
- ✨ Zaayve AI creates, inspire, and mesmerize content with brilliance!
- 📈 LinkDR gets 10x more backlinks SEO.
- ✂️ Spikes automates video editing in just 1 click.
- 🎧 Tad AI creates original songs with your choice of genres and moods using text prompts. (Featured)
- 📚 Heardly is the Fast Way to read Best Book. (Featured)
- 🦾 Flot AI writes, reads, and remembers across any apps and webs. (Featured)
- 📝 CopyOwl AI writes high-quality articles with credible sources. (Featured)
1 Million+ AI enthusiasts are eager to learn about your product.
AI Secret is the world’s #1 AI Newsletter, boasting over 1 million readers from leading companies such as OpenAI, Google, Meta, and Microsoft. We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?
What We Can Offer:
- Launch an Advertising Campaign
- Introduce New Product or Features
- Other Business Cooperation
Email our co-founder Mark directly at [email protected] if the button fails.
The UK’s Bold AI Move: “Humphrey” to the Rescue of Civil Servants
In a move that could be straight out of a satirical comedy script, the UK government has unveiled an artificial intelligence (AI) assistant named Humphrey, inspired by the cunning and manipulative Sir Humphrey Appleby from the classic sitcom Yes, Minister. While the real Sir Humphrey was known for his labyrinthine
Magna
Meta AI’s Memory: The Secret Sauce for Better Recommendations
In the ever-evolving world of artificial intelligence, Meta AI has taken a bold step forward by introducing a “memory” feature to provide more personalized recommendations. Yes, you heard that right—Meta AI now has a memory. And no, it’s not the kind of memory where it remembers to water
Magna
DeepSeek Launches Janus-Pro-7B: The New King of AI Image Generation?
The AI world has been shaken, stirred, and flipped upside down with the launch of DeepSeek’s latest multimodal AI model, the Janus-Pro-7B. This 7-billion-parameter behemoth has boldly claimed to outperform the reigning champions of text-to-image generation, OpenAI’s DALL-E 3 and Stability AI’s Stable Diffusion 3. But is it all just
Magna
Hello, AI Enthusiasts!
As we tumble down the rabbit hole of artificial intelligence, we find ourselves in a whimsical world where machines not only learn our preferences but also remember our deepest, darkest secrets—like that time you admitted to loving pineapple on pizza.
SPONSORED BY
HEADLINES
Meta AI's Memory: The Secret Sauce for Better Recommendations
Meta AI has ignited discussions with its introduction of a "memory" feature, designed to enhance user experience through personalized recommendations. This innovative capability aims to remember user-specific details, sparking both excitement and concern over privacy implications.
- Meta AI can remember personal details shared in conversations, such as dietary preferences, to tailor interactions and suggestions accordingly.
- The memory feature collects data from various Meta-owned platforms, analyzing user activity on Facebook, Instagram, and messaging apps to provide personalized recommendations and insights.
- The feature enhances user experience by providing relevant suggestions, saving time on repetitive explanations, and ensuring recommendations are closely aligned with individual preferences.
- While competitors like ChatGPT and Google Gemini offer similar personalized features, Meta AI's memory function stands out for its tailored recommendations, despite challenges with privacy.
Meta AI's Memory: The Secret Sauce for Better Recommendations
Meta AI has ignited discussions with its introduction of a "memory" feature, designed to enhance user experience through personalized recommendations. This innovative capability aims to remember user-specific details, sparking both excitement and concern over privacy implications.
TOGETHER WITH COPYOWL AI
Write High-Quality Articles With Credible Sources
CopyOwl offers a unique AI Agent that combines AI search with a long copy generator. It searches 5 to 20 relevant sources using a search engine before its AI Agent creates accurate long copy, ensuring there are no mistakes.
Key Features:
- Fully Referenced Citations: CopyOwl provides precise, academic-grade citations for every paper, enhancing trust and credibility with optional credentialed web citations for greater reliability.
- Versatile, In-Depth Content: CopyOwl delivers high-quality blogs, essays, market research, business reports, and whitepapers, all crafted with accurate insights and personalized to your unique style.
- Tailor your tone: CopyOwl lets you customize the tone and style of your content—be it academic, professional, or conversational—to perfectly align with your unique requirements.
- All in One Click: CopyOwl automates research, citations, and writing in minutes. Just input your topic and keywords to get a professional, fully referenced, and ready-to-use final copy.
HEADLINES
DeepSeek Launches Janus-Pro-7B: The New King of AI Image Generation?
The launch of DeepSeek's Janus-Pro-7B has sparked significant excitement in the AI community, positioning it as a potential challenger to established models like DALL-E 3 and Stable Diffusion 3. This new multimodal AI model, with its impressive capabilities and innovative architecture, raises questions about its true potential and the future of AI-driven creativity.
- Developed by DeepSeek AI, Janus-Pro-7B is an open-source multimodal large language model that integrates text understanding and image generation in one coherent framework, utilizing a "cross-modal attention" mechanism.
- Janus-Pro-7B outperforms its competitors in key metrics like prompt adherence (93.5% vs. DALL-E 3's 85.1%), inference speed (3.2s vs. DALL-E 3's 4.8s), and complex scene handling (89.7% vs. DALL-E 3's 84.2%).
- The model incorporates a Hybrid Transformer-Diffusion Architecture along with dynamic token routing and memory-augmented attention, addressing long-standing issues like prompt forgetting.
- Janus-Pro-7B was trained on the DeepSeek-Vision Corpus, comprising 1.2 billion high-quality text-image pairs, and utilizes 400 million synthetic prompts for edge case training.
- Using proprietary SparQ optimization, Janus-Pro-7B achieves a 65% reduction in energy costs compared to Stable Diffusion 3, showcasing an eco-friendly approach to AI development.
DeepSeek Launches Janus-Pro-7B: The New King of AI Image Generation?
The launch of DeepSeek's Janus-Pro-7B has sparked significant excitement in the AI community, positioning it as a potential challenger to established models like DALL-E 3 and Stable Diffusion 3. This new multimodal AI model, with its impressive capabilities and innovative architecture, raises questions about its true potential and the future of AI-driven creativity.
QUICK HITS
- DeepSeek announces it is limiting registrations due to a large-scale cyberattack.
- US nuclear fusion start-up Helion, backed by Sam Altman and Peter Thiel, raises $425 million to advance its electricity production goals by 2028.
- DeepSeek's mobile app has overtaken ChatGPT to become the top app in the U.S. App Store and 51 other countries.
- France's AI chatbot Lucie, backed by President Macron, has been suspended shortly after launch due to widespread inaccuracies.
- President Trump announced that Microsoft is in talks to acquire TikTok, urging for a "bidding war" over the sale.
TRENDING TOOLS
- 📜 AutoCaret creates documents faster by talking to AI.
- ♨️ Meals AI creates amazing recipes.
- 🎬 Cutlabs.ai helps streamers make highlight clips.
- ✂️ Foto AI swiftly removes and replaces image backgrounds.
- 🗪 Notionsmith can browse and understand the web.
- 🎧 Tad AI creates original songs with your choice of genres and moods using text prompts. (Featured)
- 📚 Heardly is the Fast Way to read Best Book. (Featured)
- 🦾 Flot AI writes, reads, and remembers across any apps and webs. (Featured)
- 🗃️ UnDatas transforms unstructured data into AI ready assets. (Featured)
1 Million+ AI enthusiasts are eager to learn about your product.
AI Secret is the world’s #1 AI Newsletter, boasting over 1 million readers from leading companies such as OpenAI, Google, Meta, and Microsoft. We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?
What We Can Offer:
- Launch an Advertising Campaign
- Introduce New Product or Features
- Other Business Cooperation
Email our co-founder Mark directly at [email protected] if the button fails.
Unveiling the DeepSeek Creator: Hedge Fund Billionaire Liang Wenfeng
In the world of artificial intelligence (AI), where Silicon Valley giants like OpenAI and Google have long dominated the headlines, a new name has emerged from the East to challenge the status quo: Liang Wenfeng. The 40-year-old hedge fund manager who founded DeepSeek, a Chinese AI startup, has not only
MagnaPerplexity AI Proposes TikTok Merger with 50% US Ownership: A Groundbreaking Proposal or Chaos?
Ladies and gentlemen, gather around because we have a tech drama that could rival the plot twists of your favorite Netflix series! In a move that has left tech enthusiasts, policymakers, and TikTok dancers scratching their heads (and maybe doing a celebratory shuffle), Perplexity AI has proposed a merger with
Magna
Stocks Fallen: Nvidia hit by DeepSeek
Ladies and gentlemen, gather around for the latest episode of “Tech Wars: The Rise of the AI Titans.” Today’s protagonist—or antagonist, depending on which side of the stock market you’re on—is none other than China’s DeepSeek AI. This scrappy startup has managed to shake the
Magna
Hello, AI Enthusiasts!
In today’s delightful dish of digital drama, we have Perplexity AI serving up a proposal that’s juicier than your morning gossip. With the U.S. government potentially owning up to 50% of this new entity – but without any voting rights, because who wants to ruin a good thing with actual say-so?
SPONSORED BY
HEADLINES
Perplexity AI Proposes TikTok Merger with 50% US Ownership: A Groundbreaking Proposal or Chaos?
In a bold and controversial proposal, Perplexity AI has suggested a merger with TikTok's U.S. operations, potentially leading to the U.S. government owning a significant stake in the new entity. This move raises questions about national security, corporate governance, and the cultural implications of government involvement in social media.
- Perplexity AI has revised its merger proposal with ByteDance, aiming to create a new entity that combines its operations with TikTok's U.S. business, with the U.S. government potentially owning up to 50% after an IPO valued at $300 billion.
- This proposal arises amid longstanding concerns over TikTok's Chinese ownership, particularly regarding data security and national sovereignty, especially given the recent law banning TikTok in the U.S.
- The proposed IPO's staggering valuation of $300 billion signifies substantial economic implications, with the U.S. government holding a significant stake but without voting rights, and ByteDance retaining control over its proprietary algorithm.
Perplexity AI Proposes TikTok Merger with 50% US Ownership: A Groundbreaking Proposal or Chaos?
In a bold and controversial proposal, Perplexity AI has suggested a merger with TikTok's U.S. operations, potentially leading to the U.S. government owning a significant stake in the new entity. This move raises questions about national security, corporate governance, and the cultural implications of government involvement in social media.
ByteDance: Perplexity AI Proposes TikTok Merger with 50% US Ownership: A Groundbreaking Proposal or Chaos?
Perplexity AI has revised its merger proposal with ByteDance, aiming to create a new entity that combines its operations with TikTok's U.S. business, with the U.S. government potentially owning up to 50% after an IPO valued at $300 billion.
TOGETHER WITH CHATHUB
Compare Multiple AI Chatbots At Once
Struggling to choose between countless AI models and tools? ChatHub simplifies your workflow by allowing you to compare responses from multiple chatbots simultaneously.
Ideal for advanced users who need reliable, creative, and accurate outputs, ChatHub saves time by offering cross-chatbot validation and prompt templates, ensuring you get the best results without switching between apps.
HEADLINES
Stocks Fallen: Nvidia hit by DeepSeek
In the rapidly evolving landscape of artificial intelligence, the emergence of China’s DeepSeek AI has sent shockwaves through the tech industry, particularly affecting established giants like Nvidia.
- DeepSeek’s AI models are 20-40 times cheaper than those from OpenAI, offering a cost-effective alternative that challenges the pricing strategies of established AI companies.
- Nvidia’s stock plummeted by 14% in response to DeepSeek's success, prompting analysts to question the value of premium chips in light of more affordable alternatives that achieve similar results.
- DeepSeek's rise signifies a shift in AI economics, showcasing that high-quality AI development doesn't necessarily require massive funding and raising geopolitical concerns about tech competition, particularly between the U.S. and China.
Stocks Fallen: Nvidia hit by DeepSeek
Nvidia’s stock plummeted by 14% in response to DeepSeek's success, prompting analysts to question the value of premium chips in light of more affordable alternatives that achieve similar results.
Stocks Fallen: Nvidia hit by DeepSeek
DeepSeek's rise signifies a shift in AI economics, showcasing that high-quality AI development doesn't necessarily require massive funding and raising geopolitical concerns about tech competition, particularly between the U.S. and China.
QUICK HITS
- AI recruiting startup Mercor secures $75 million in funding, valuing the company at $2 billion.
- Apple has appointed veteran exec Kim Vorrath to lead its AI initiatives and revamp Siri.
- ElevenLabs raises $250 million in Series C funding, reaching a $3 billion valuation amid growing competition in AI voice technology.
- Qwen Team announces the release of open-source Qwen2.5-1M models, supporting context lengths of up to 1 million tokens.
- Mark Zuckerberg announced a $60-$65 billion investment in AI infrastructure for 2025 to compete with OpenAI and Google.
TRENDING TOOLS
- 🏃♀️ AI Runner provides custom AI coaching, heart rate tracking, and instant feedback.
- 🔜 PDF2Document is a powerful PDF to Word converter.
- 📑 GitBook builds Docs as a product.
- 🎶 FineShare makes it easy to create realistic voices.
- 🦾 Dust builds custom AI assistants to speed up your work.
- 🎧 Tad AI creates original songs with your choice of genres and moods using text prompts. (Featured)
- 📚 Heardly is the Fast Way to read Best Book. (Featured)
- 🦾 Flot AI writes, reads, and remembers across any apps and webs. (Featured)
- 📝 CopyOwl AI writes high-quality articles with credible sources. (Featured)
- 🏹 DreamStori allows you to experience romance through interactive stories. (Featured)
1 Million+ AI enthusiasts are eager to learn about your product.
AI Secret is the world’s #1 AI Newsletter, boasting over 1 million readers from leading companies such as OpenAI, Google, Meta, and Microsoft. We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?
What We Can Offer:
- Launch an Advertising Campaign
- Introduce New Product or Features
- Other Business Cooperation
Email our co-founder Mark directly at [email protected] if the button fails.
The Explosive Popularity of DeepSeek: A Chip Investigation Waiting to Happen?
Join the Top1 AI Newsletter with 1.5M readers. Learn to leverage AI in your Business and Career before everyone else - Join Free↓
Magna
The Stargate Project: A $500 Billion AI Odyssey
In a move that feels like it was ripped straight from a sci-fi blockbuster, OpenAI, SoftBank, Oracle, and other tech titans have joined forces to launch “The Stargate Project.” This ambitious initiative, announced on January 21, 2025, is set to pump a jaw-dropping $500 billion into U.S. artificial intelligence
Magna
How would you feel if someone could locate you just seconds after you post a picture?
Imagine this: you upload a photo of your last vacation to social media, and within seconds, someone can pinpoint the exact location where it was taken—not from metadata, but just by analyzing the image itself. Sounds like something out of a sci-fi thriller, doesn’t it? Well, welcome to
Magna
Ladies and gentlemen, gather around for the latest episode of "Tech Wars: The Rise of the AI Titans." Today’s protagonist—or antagonist, depending on which side of the stock market you’re on—is none other than China’s DeepSeek AI. This scrappy startup has managed to shake the foundations of Silicon Valley and send Nvidia, the reigning king of AI hardware, into a tailspin. If you thought AI was just about chatbots and self-driving cars, think again. DeepSeek has made it clear that the battle for AI supremacy is also a battle for cost efficiency, technological dominance, and, apparently, Wall Street’s collective sanity.
Stocks Fallen: Nvidia hit by DeepSeek
Ladies and gentlemen, gather around for the latest episode of "Tech Wars: The Rise of the AI Titans." Today’s protagonist—or antagonist, depending on which side of the stock market you’re on—is none other than China’s DeepSeek AI.
A Cinderella Story (With Chips Instead of Glass Slippers)
DeepSeek’s models are reportedly 20-40 times cheaper than equivalent models from OpenAI. Yes, you read that right. While OpenAI is out here charging premium prices, DeepSeek is essentially running a "Black Friday Sale" on AI models year-round.
Unlike most AI giants that rely on Nvidia’s high-performance chips, DeepSeek has managed to achieve impressive results using less powerful hardware. This is like winning a Formula 1 race in a Prius—unheard of but undeniably impressive.
DeepSeek’s models are open-source, which means they’re available for free. This has not only democratized access to advanced AI but also raised questions about whether the tech industry has been overspending on proprietary solutions.
DeepSeek’s R1 model has skyrocketed to the top of Apple’s App Store charts, dethroning OpenAI’s ChatGPT as the most downloaded free app. This is like a new band knocking The Beatles off the charts—an event that’s sure to ruffle some feathers.
OpenAI: A Cinderella Story (With Chips Instead of Glass Slippers)
DeepSeek’s R1 model has skyrocketed to the top of Apple’s App Store charts, dethroning OpenAI’s ChatGPT as the most downloaded free app. This is like a new band knocking The Beatles off the charts—an event that’s sure to ruffle some feathers.
Apple: A Cinderella Story (With Chips Instead of Glass Slippers)
DeepSeek’s R1 model has skyrocketed to the top of Apple’s App Store charts, dethroning OpenAI’s ChatGPT as the most downloaded free app. This is like a new band knocking The Beatles off the charts—an event that’s sure to ruffle some feathers.
Nvidia’s Nightmare: The Fallout
Now, let’s talk about Nvidia, the chip-making giant that has been the backbone of AI development for years. Nvidia’s GPUs are the gold standard for training and running large language models (LLMs). However, DeepSeek’s success has thrown a wrench into Nvidia’s well-oiled machine. Here’s how:
Nvidia’s stock plunged by as much as 14% in pre-market trading on January 27, 2025. Other tech stocks, including ASML and Broadcom, also took a hit, with the Nasdaq index dropping by nearly 4%. It’s safe to say that Wall Street is not having a good day.
Analysts are now questioning whether Nvidia’s premium chips are still worth the price. If companies like DeepSeek can achieve similar results with less powerful hardware, why pay a premium? This has led to a broader reckoning about the economics of AI.
DeepSeek’s ability to train models using less powerful chips could inspire other companies to follow suit. This would reduce demand for Nvidia’s high-end GPUs, potentially impacting the company’s revenue and market dominance .
Nvidia’s Nightmare: The Fallout
Now, let’s talk about Nvidia, the chip-making giant that has been the backbone of AI development for years. Nvidia’s GPUs are the gold standard for training and running large language models (LLMs). However, DeepSeek’s success has thrown a wrench into Nvidia’s well-oiled machine. Here’s how:
Implications for the Tech Industry
DeepSeek’s rise is not just a story about one company disrupting another; it’s a wake-up call for the entire tech industry. Here are some of the broader implications:
- Shift in AI Economics: DeepSeek has proven that you don’t need to spend billions to develop cutting-edge AI. This could lead to a shift in how companies allocate resources and prioritize efficiency over brute force.
- Geopolitical Ramifications: The U.S. government has been trying to maintain its lead in AI by restricting the export of advanced chips to China. However, DeepSeek’s success shows that China can innovate even with limited access to high-end hardware. This could have significant implications for global tech competition.
- Open Source vs. Proprietary Models: DeepSeek’s open-source approach challenges the proprietary models of companies like OpenAI and Google. This could democratize access to AI but also raise questions about intellectual property and data security.
What’s Next for Nvidia?
So, what does the future hold for Nvidia? While the company is undoubtedly facing challenges, it’s not time to write its obituary just yet. Nvidia still has a significant advantage in high-end GPUs, which are essential for many advanced applications. However, the company will need to adapt to the changing landscape by focusing on cost efficiency and exploring new markets.
In the grand scheme of things, DeepSeek’s rise is a testament to the power of innovation and the importance of challenging the status quo. While Nvidia may be feeling the heat, this disruption could ultimately lead to a more competitive and efficient AI industry. As for DeepSeek, it has proven that sometimes, the underdog can not only compete but also redefine the game.
So, here’s to DeepSeek for shaking things up and to Nvidia for hopefully rising to the challenge. And to the rest of us, let’s grab some popcorn and enjoy the show—because the AI wars are just getting started.
Hello, AI Enthusiasts!
Today, we turn our attention to DeepSeek, the enigmatic powerhouse from the East that has everyone buzzing more than a caffeinated swarm of drones. Hot on the heels of their latest innovations, the only things they seem to be generating faster than cutting-edge AI are rumors and doubts. Who knew AI could be so… dramatic?
SPONSORED BY
HEADLINES
The Explosive Popularity of DeepSeek: A Chip Investigation Waiting to Happen?
DeepSeek, a Chinese AI startup that is shaking up the landscape of artificial intelligence and challenging U.S. dominance in the field. With a groundbreaking AI model that performs exceptionally well using less advanced hardware, DeepSeek's success raises questions about U.S. export controls and the future of AI innovation.
- DeepSeek has gained attention with its AI model, DeepSeek-R1, which outperforms models from U.S. tech giants on key benchmarks, achieving a 90.2% exact match rate on the MATH 500 benchmark.
- Despite tightening restrictions from the U.S. on advanced chips, DeepSeek has thrived, demonstrating that such measures may not effectively stifle innovation in China, but rather drive efficiency and collaboration.
- DeepSeek-R1 has shown superior performance in mathematical reasoning and coding tasks at a fraction of the cost, but its capabilities in broader applications remain to be evaluated.
- Allegations suggest that DeepSeek may have utilized outputs from existing AI models like ChatGPT, raising ethical questions about intellectual property and the fairness of AI development.
The Explosive Popularity of DeepSeek: A Chip Investigation Waiting to Happen?
DeepSeek, a Chinese AI startup that is shaking up the landscape of artificial intelligence and challenging U.S. dominance in the field. With a groundbreaking AI model that performs exceptionally well using less advanced hardware, DeepSeek's success raises questions about U.S. export controls and the future of AI innovation.
TOGETHER WITH PIXNO
Turn Photos Into Notes With AI
Ladies and gentlemen, tech enthusiasts, procrastinators, and anyone whose idea of organization is a mystical blend of sticky notes and sheer chaos—allow us to introduce you to the future of note-taking: Pixno!
Picture this: You snap a photo of a complex chart that looks like it was designed by a mad scientist. Instead of spending hours trying to decipher it, Pixno swoops in with its AI-Enhanced Clarity feature. It’s like having a personal assistant who knows exactly what you mean, even when you don’t. With a wave of its virtual wand (or maybe just some clever algorithms), Pixno transforms your intricate visuals into well-structured notes that are as easy to read as a toddler's bedtime story.
HEADLINES
Explore DeepSeek’s Key Innovations: Data Distillation Tech and MoE Architecture
The introduction of DeepSeek V3 signifies a groundbreaking evolution in the realm of artificial intelligence, showcasing how advanced architectures and efficient training methodologies can lead to unparalleled performance in large language models (LLMs). This model not only pushes the boundaries of what AI can achieve but also highlights the importance of cost-efficiency and scalability in its design.
- DeepSeek V3 uses data distillation to improve and shrink training datasets, which boosts model performance by selecting high-quality tokens and allowing for efficient pre-training. It requires just 2.788 million GPU hours. This method also ensures stability and strength during the training process.
- At its core, DeepSeek V3 employs a MoE architecture that enables efficient scaling by activating only a subset of experts relevant to the specific task, significantly reducing computational overhead. A gating network determines which experts to activate for each input query, optimizing both performance and efficiency by engaging only the most pertinent components.
- DeepSeek V3 excels in various benchmarks, achieving state-of-the-art results in multi-task language understanding, mathematical reasoning, and coding tasks, highlighting its effectiveness.
- Despite its advancements, challenges such as expert imbalance and scalability limits remain, necessitating further research into efficient gating mechanisms and cross-expert communication improvements.
Explore DeepSeek’s Key Innovations: Data Distillation Tech and MoE Architecture
The introduction of DeepSeek V3 signifies a groundbreaking evolution in the realm of artificial intelligence, showcasing how advanced architectures and efficient training methodologies can lead to unparalleled performance in large language models (LLMs). This model not only pushes the boundaries of what AI can achieve but also highlights the importance of cost-efficiency and scalability in its design.
QUICK HITS
- Paul McCartney warns that proposed changes to copyright law could enable AI to exploit artists, risking their livelihoods.
- The US Forest Service is using AI to investigate the causes of recent wildfires in Los Angeles, as over half of such fires in the Western US remain unexplained.
- The Stargate will utilize solar and battery power for its AI data centers.
- OpenAI is facing scrutiny after it was revealed that it helped design the math test that its o3 model excelled in.
- The use of AI in the film "The Brutalist" sparks controversy amid the 2025 Oscar nominations.
TRENDING TOOLS
- 🎥 Wisecut turns long videos into highlights.
- 🏷️ Logo Diffusion designs your Logo with AI.
- ⌨ Cursor is an AI-first code editor.
- 🎯 Venturekit writes a business plan for you.
- 🛎️ FindAMeal uses AI to find your next favourite restaurant.
- 🎧 Tad AI creates original songs with your choice of genres and moods using text prompts. (Featured)
- 📚 Heardly is the Fast Way to read Best Book. (Featured)
- 🦾 Flot AI writes, reads, and remembers across any apps and webs. (Featured)
- 📝 CopyOwl AI writes high-quality articles with credible sources. (Featured)
1 Million+ AI enthusiasts are eager to learn about your product.
AI Secret is the world’s #1 AI Newsletter, boasting over 1 million readers from leading companies such as OpenAI, Google, Meta, and Microsoft. We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?
What We Can Offer:
- Launch an Advertising Campaign
- Introduce New Product or Features
- Other Business Cooperation
Email our co-founder Mark directly at [email protected] if the button fails.
Leaked documents show Microsoft Strengthens Ties with Israeli Military for AI Support in Gaza War
In a world where technology is supposed to bring us closer together, it seems like some tech giants have been busy playing war games instead. Microsoft, the tech behemoth we all know for Windows updates that always come at the worst time, has found itself under the spotlight for its
Magna
OpenAI Launches Operator: An AI Agent Transforming Task Management
On January 23, 2025, OpenAI unveiled “Operator,” its groundbreaking AI agent designed to autonomously perform tasks on behalf of users. This release marks a significant milestone in the evolution of artificial intelligence, transitioning from conversational AI models like ChatGPT to action-oriented agents capable of directly interacting with web browsers and
Magna
A Claude-tastic Future: Web Access, Rate Limits, and New Models Are Coming Your Way!
Joanna Stern, the Wall Street Journal’s Senior Personal Technology Columnist, just wrapped up a fascinating tête-à-tête with Anthropic CEO Dario Amodei at Davos. For those of us who have been eagerly awaiting upgrades to Claude, Anthropic’s flagship AI model, the wait is almost over. Web access, improved rate
Magna
5 min read · Members
Microsoft‘s New FP4 Training Framework vs. DeepSeek: Which Training Cost is Lower?
Microsoft has just dropped a bombshell in the world of artificial intelligence (AI) and machine learning (ML). With the launch of the first-ever FP4 training framework,
6 min read
🛎️Your Toaster Might Run for Congress
Comparison Of Alibaba's New Qwen2.5-Max, DeepSeek-V3, GPT-4o, Claude-3.5-Sonnet, And Llama-3.1-405B
6 min read · Members
Alibaba Claims Its New Qwen Model Outperforms DeepSeek
Comparison Of Qwen2.5-Max, DeepSeek-V3, GPT-4o, Claude-3.5-Sonnet, And Llama-3.1-405B
6 min read
🛎️Meta Knows You More Now
The New King of AI Image Generation is Rising.
4 min read · Members
DeepSeek Launches Janus-Pro-7B: The New King of AI Image Generation?
The AI world has been shaken, stirred, and flipped upside down with the launch of DeepSeek's latest multimodal AI model, the Janus-Pro-7B. This 7-billion-parameter
4 min read · Members
Unveiling the DeepSeek Creator: Hedge Fund Billionaire Liang Wenfeng
In the world of artificial intelligence (AI), where Silicon Valley giants like OpenAI and Google have long dominated the headlines, a new name has emerged from
5 min read
🛎️A Convenient Marriage
Perplexity AI Proposes TikTok Merger with 50% US Ownership
4 min read
Stocks Fallen: Nvidia hit by DeepSeek
Ladies and gentlemen, gather around for the latest episode of "Tech Wars: The Rise of the AI Titans." Today’s protagonist—or antagonist, depending
6 min read
🛎️AI Dramatic
The Explosive Popularity of DeepSeek: A Chip Investigation Waiting to Happen?