With the seismic impact of DeepSeek on AI, the stock market, and geopolitics, we wanted to follow-up our previous post with a deeper exploration of the topic. In this post, we found 5 videos that will help you get up to speed on the unfolding drama.
Vid1: CNBC Covers the Ensuing Market Meltdown
CNBC discusses the impact of China's new AI model, DeepSeek, on the global tech industry. DeepSeek's superior efficiency and performance, even surpassing some American models, is causing a major sell-off in AI-related stocks, particularly impacting companies like Nvidia. The video explores concerns about DeepSeek's potential access to advanced technology and the implications for US technological dominance. The discussion also touches upon the shift towards open-source AI models and the uncertainty surrounding future investments in AI development. Finally, the video highlights the rapid advancement of AI technology and its potential societal impact, comparing the situation to the Sputnik moment of the space race.
Vid2: AI Enthusiast, Matt Wolfe, Gives His Take
Matt Wolfe, who closely follows the AI space, discusses DeepSeek R1, a new Chinese
open-source AI model that has caused significant market reactions. DeepSeek's
impressive performance, achieved with significantly less computing power
than comparable models like GPT-4, is attributed to its efficient training
methods and innovative design. Controversy surrounds DeepSeek's claims regarding
its resource usage, with some suggesting the company downplayed the actual
computational resources employed. Despite this, the video argues the model's
impact may be positive, possibly lowering the barrier to entry for AI
development and increasing overall demand for GPUs. The video also
covers DeepSeek's image generation model, Janice Pro 7B, and provides
instructions on how to access and utilize DeepSeek.
Vid3: A Geopolitical Perspective on the DeepSeek Saga
Here is Cold Fusion’s take on the DeepSeek story. He
discusses the sudden emergence of DeepSeek R1, a free, open-source Chinese AI
model that rivals—and in some ways surpasses—leading American AI models. Its
unexpectedly low development cost and superior efficiency have sent shockwaves
through the US stock market and prompted a reassessment of AI development
strategies. Concerns about intellectual property theft are raised, alongside
geopolitical implications of this technological advancement. The narrative
explores the innovative techniques behind DeepSeek R1's performance and the
competitive landscape it has created, highlighting the resulting cost
reductions and potential for rapid AI progress globally.
Vid4: If you are using DeepSeek, Your Data is Going to China!
Skill Leap AI discusses serious privacy concerns regarding the
DeepSeek website and app, highlighting issues like vague data retention
policies, data storage in China raising compliance issues with international
laws, lack of transparency in data usage, and insufficient age verification.
The creator outlines these issues after reviewing the platform's privacy policy
and terms of service using ChatGPT. To mitigate these risks, the video suggests
using locally installed versions of DeepSeek R1 or utilizing DeepSeek's
integration within the PerplexityAI search engine, a US-based service. Finally,
the video promises a future comparison of DeepSeek R1 and ChatGPT's 01 model.
Vid5: A Video Walkthrough of Dario Amodei's take on DeepSeek's Capabilities
In this video, Matt Berman takes a look at Dario Amodei's
take on the DeepSeek saga. Amodei, the current CEO of OpenAI’s chief rival Anthropic,
wrote an essay discussing the implications of DeepSeek's AI model, R1,
particularly concerning its potential data acquisition from OpenAI and the
resulting impact on the AI industry and geopolitical landscape. The essay
analyzes the three key dynamics of AI development: scaling laws, the
shifting curve, and paradigm shifts, emphasizing the escalating costs and
exponential advancements in AI capabilities. Concerns about China's
access to advanced GPUs and their potential to achieve artificial
general intelligence (AGI) are also highlighted, underscoring the importance of
export controls. Finally, the essay argues that DeepSeek's cost-effective
model, while impressive, does not represent a fundamental shift in AI economics
and that the market's overreaction was unwarranted.
Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a Sr. AI Product Manager who is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else. This post was written with the assistance of an AI language model. The model provided suggestions and completions to help me write, but the final content and opinions are my own.
In my ongoing exploration of emerging AI technologies, I recently encountered a fascinating CNBC deep dive into DeepSeek, a Chinese AI firm that's reshaping the competitive landscape of artificial intelligence. The investigation sparked my interest in understanding how this relatively unknown company has managed to develop AI models that rival industry giants like OpenAI and Google, while maintaining an open-source approach and significantly lower development costs. Through comprehensive research and analysis of DeepSeek's technical innovations, particularly their DeepSeek-V3 and DeepSeek-R1 models, I've uncovered insights into how this disruptor is challenging conventional wisdom about AI development costs and accessibility. This article examines DeepSeek's technological breakthroughs, their implications for the global AI race, and what their emergence means for making advanced AI accessible to all.
Intro
DeepSeek, a relatively unknown Chinese AI firm with roots in the quantitative stock trading firm High-Flyer1 has sent ripples through the AI community with its release of DeepSeek-V3 and DeepSeek-R1, two powerful open-source AI systems. These models are impressive not only for their technical capabilities, which rival those of industry giants like OpenAI and Google, but also for their remarkably low development costs and open-source accessibility. This has sparked considerable discussion about the evolving AI landscape and the intensifying competition between the US and China in AI development.
DeepSeek-V3: A Technical Marvel
With a massive 671 billion parameters, DeepSeek-V3 surpasses even Meta's Llama 3.1 in scale2. However, DeepSeek-V3 distinguishes itself through its innovative Mixture-of-Experts (MoE) architecture. This architecture activates only the necessary neural networks for specific tasks, resulting in significant cost savings and improved efficiency3. Despite its vast number of parameters, DeepSeek-V3 operates with just 37 billion during actual tasks3. This efficient design allows it to achieve high performance with significantly less computational power and cost compared to its peers4.
DeepSeek-V3 excels in various text-based tasks, including coding, translation, and writing2. It has achieved top scores on popular AI benchmarks, such as HumanEval, GSM8K, and MMLU, challenging both open and closed-source models3. Notably, DeepSeek-V3 outperformed Meta's Llama 3.1, OpenAI's GPT-4o, and Anthropic's Claude Sonnet 3.5 in accuracy across various tasks, from complex problem-solving to math and coding6.
However, it's important to acknowledge that DeepSeek-V3 is primarily focused on text-based tasks and does not possess multimodal abilities2. Like many large language models, it may also inherit biases from its training data, requiring careful consideration in real-world applications2.
DeepSeek-R1: Mastering Reasoning
DeepSeek further solidified its position with the release of DeepSeek-R1, a reasoning model designed to tackle complex problems with a focus on logical inference, mathematical problem-solving, and real-time decision-making7. This sets it apart from traditional language models, which primarily focus on text generation and comprehension.
DeepSeek-R1's development began with DeepSeek-R1-Zero, a foundational model trained exclusively via reinforcement learning8. While R1-Zero showed promise in reasoning, it faced challenges with readability and output coherence. DeepSeek addressed these issues in R1 by incorporating cold-start data and a multi-stage reinforcement learning process9.
DeepSeek-R1 has demonstrated remarkable performance on various benchmarks, including AIME (American Invitational Mathematics Examination) and MATH1. While DeepSeek claimed that R1 exceeded the performance of OpenAI's o1 on these benchmarks, independent analysis by The Wall Street Journal found that o1 was faster in solving AIME problems1. Nevertheless, R1's performance remains competitive with leading models in the field.
Feature
DeepSeek-R1
OpenAI o1
Architecture
Mixture-of-Experts (MoE)
-
Parameters
671 billion total, 37 billion active
-
AIME 2024 (Pass@1)
79.8%
79.2%
MATH-500 (Pass@1)
97.3%
96.4%
Codeforces (Percentile)
96.3
96.6
Cost (per million tokens)
$2.19
$60
Key Features
Open-source, transparent reasoning
Chain-of-thought processing
DeepSeek-R1 has also shown promising results in financial analysis, outperforming the S&P 500 and maintaining superior Sharpe and Sortino ratios compared to the market7. Furthermore, it exhibits a unique ability to provide transparent reasoning, offering insights into its decision-making process10. However, it's worth noting that the model tends to align with the official Chinese government position on sensitive political topics10.
DeepSeek-VL: Expanding into Vision-Language
Beyond its language models, DeepSeek is also exploring vision-language (VL) capabilities with DeepSeek-VL11. This model is designed for real-world vision and language understanding applications, with the ability to process logical diagrams, web pages, formulas, scientific literature, and natural images11. DeepSeek-VL showcases the company's commitment to advancing AI research across multiple modalities.
Open-Source and Cost-Effective: AI For All
One of the most significant aspects of DeepSeek's models is their open-source nature. This allows developers and researchers to freely access, modify, and deploy the models, fostering collaboration and innovation within the AI community4. This open approach contrasts with the proprietary models of many US-based companies and has the potential to democratize access to advanced AI technologies12.
DeepSeek has achieved these impressive results with significantly lower development costs. DeepSeek-V3 was reportedly trained in around 55 days at a cost of US$5.58 million, using considerably fewer resources compared to its peers1. This cost-effectiveness challenges the existing paradigm in the AI industry, where high performance has typically been associated with high costs4.
Implications for the US-China AI Race
DeepSeek's emergence has raised concerns in the US about China's growing AI capabilities13. The US government has implemented restrictions on China's access to advanced AI chips, aiming to curb its progress in AI development14. However, DeepSeek has demonstrated that Chinese researchers can develop world-class AI models with limited resources and by leveraging open-source technologies15. This raises questions about the effectiveness of these restrictions and their potential to inadvertently spur innovation in China by forcing researchers to focus on efficiency and alternative approaches14.
DeepSeek's success has been attributed to several factors, including its efficient MoE architecture, innovative training methods, and a focus on maximizing resource utilization16. The company's ability to develop high-performing models at a fraction of the cost of its US counterparts has put pressure on companies like Meta and OpenAI to rethink their AI strategies17.
Expert Opinions and Analysis
Experts in the AI field have recognized DeepSeek's significant contributions to AI development. They highlight the model's impressive performance, cost-effectiveness, and open-source nature as key factors that could reshape the AI landscape7. Some experts suggest that DeepSeek's approach could lead to a democratization of AI, making advanced AI capabilities more accessible to a wider range of developers and researchers12. Others emphasize the potential for DeepSeek to accelerate innovation and competition in the AI industry, potentially leading to breakthroughs in various fields18.
Use Cases and Applications
DeepSeek's models have shown potential in a variety of applications. DeepSeek-R1, for example, has been used to run complex reasoning tasks on smartphones, generate code for rotating objects with collision detection, and even build a clone of the AI-powered conversational search engine Perplexity AI19. The models have also shown promise in areas like software development, business operations, and education3.
Synthesis
DeepSeek's emergence as a major player in the AI landscape has significant implications for the future of AI development. Its open-source approach, combined with its high-performing and cost-effective models, challenges the dominance of US-based companies and has the potential to democratize access to advanced AI technologies. This could foster a more diverse and inclusive AI ecosystem, with wider participation from developers and researchers worldwide.
DeepSeek's success also highlights the growing capabilities of Chinese AI research and the potential for open-source technologies to disrupt the AI industry. As the AI race intensifies, DeepSeek's innovative approach and commitment to accessibility could shape the future of AI development and its impact on the global technology landscape.
Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a Sr. AI Product Manager who is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else. This post was written with the assistance of an AI language model. The model provided suggestions and completions to help me write, but the final content and opinions are my own.
Gemini 2.0: Google’s Free Model Outshines OpenAI in Science and Math
Google has unveiled Gemini 2.0 Flash Thinking, a cutting-edge AI model that rivals OpenAI's premium offerings by being both advanced and free during beta testing. With standout features like million-token processing capabilities and enhanced reasoning transparency, the model sets benchmarks in mathematics and science tasks, outperforming previous iterations and competitors. Additionally, Gemini 2.0 integrates native code execution for direct programming within the system and boasts improved safeguards against contradictions. Industry analysts suggest its accessibility and transparency could redefine AI development, challenging OpenAI's dominance and making the technology more approachable for developers and researchers worldwide.
Performance & Transparency: Gemini 2.0 excels in advanced tasks and reveals its reasoning process, addressing AI’s "black box" problem.
Million-Token Context Window: The model processes vast datasets simultaneously, enabling breakthroughs in research and analytics.
Strategic Release: Google's free beta version may attract users away from OpenAI's premium $200 subscription.
Stargate: $500 Billion AI Partnership Set to Transform U.S. Economy
President Donald Trump has announced the launch of Stargate, a $500 billion AI infrastructure initiative spearheaded by OpenAI, Oracle, and SoftBank. Initially starting with a $100 billion investment in Texas, the project aims to build data centers and energy facilities to support AI development. Trump emphasized the partnership's potential to transform the U.S. economy, while leaders like Masayoshi Son (SoftBank), Sam Altman (OpenAI), and Larry Ellison (Oracle) hailed the initiative as a defining project for this era. Though initiated under the Biden administration, Stargate signals the U.S.'s commitment to leading AI innovation amidst global competition.
Massive Investment: Stargate’s $500 billion funding underscores the importance of AI infrastructure in the U.S. economy.
Strategic Partnerships: OpenAI, Oracle, and SoftBank are pooling resources to develop cutting-edge data and energy infrastructure.
AI Leadership: The project reflects America’s ambition to stay ahead in the global AI race, particularly against China.
Trump Pardons Silk Road Founder Ross Ulbricht, Sparking Debate
President Donald Trump has pardoned Ross Ulbricht, the founder of the infamous Silk Road marketplace, who was serving a life sentence for facilitating online drug sales and other illicit activities. Ulbricht, known by his pseudonym “Dread Pirate Roberts,” created Silk Road in 2011, a platform that operated on the dark web using cryptocurrency. Though prosecutors linked the site to drug overdose deaths and murder-for-hire conspiracies, Ulbricht denied responsibility for user actions on the platform. Trump cited Ulbricht’s case as an example of government overreach, with the pardon appealing to libertarian and cryptocurrency communities advocating for his release.
Silk Road's Infamy: The dark web marketplace facilitated over $200 million in illegal transactions and was shut down in 2013.
Pardon’s Appeal: Trump’s pardon highlights his alignment with libertarian voters and cryptocurrency supporters.
Ongoing Debate: Critics argue about justice for Ulbricht versus the ethical and legal implications of his actions.
Biden’s AI Risk Order Revoked: A Shift Towards Deregulation
President Donald Trump has repealed a 2023 executive order issued by Joe Biden that mandated stricter oversight of AI development to address risks to national security, public safety, and the economy. Biden’s order required developers of high-risk AI systems to submit safety test results to the federal government and established standards for testing AI systems for potential threats. The Trump administration argued that these measures hindered AI innovation and removed them in favor of promoting free-market-driven AI development. Critics fear that the repeal could exacerbate AI’s risks, such as cybersecurity vulnerabilities and ethical concerns, without proper safeguards.
Policy Reversal: Biden’s AI safety measures were removed, prioritizing rapid AI growth over regulatory oversight.
Innovation vs. Risk: The repeal raises concerns about unchecked AI development potentially leading to safety and security threats.
Partisan Divide: Highlights differing approaches to AI regulation, with Trump’s focus on deregulation.
OpenAI’s Operator: The Future of AI-Driven Task Automation
OpenAI is set to launch Operator, a "computer use agent" that can perform tasks directly in a user’s browser, representing a significant step toward AI-driven task automation. Operator will assist users by navigating online platforms, performing tasks like finding flights or drafting emails. While the system relies on multimodal AI to analyze text and visuals, it keeps users involved in critical steps, such as completing transactions. Despite its potential, Operator raises concerns about misuse, like spamming and bypassing restrictions, and could encounter reliability issues similar to early self-driving cars. OpenAI’s push for this functionality reflects the broader race to create general AI capable of replacing human workflows.
AI for Automation: Operator enables task automation, such as navigating websites and performing online actions.
Ethical Concerns: Potential misuse, like spamming and data privacy risks, highlights the need for robust safeguards.
AI Evolution: The feature represents a step toward achieving artificial general intelligence by bridging productivity gaps.
Update: OpenAI Unveiled "Operator" in the following livestream on Jan 23, 2024 (just after this post was published):
Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a Sr. AI Product Manager who is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else. This post was written with the assistance of an AI language model. The model provided suggestions and completions to help me write, but the final content and opinions are my own.
Alexa’s Overhaul: Tackling AI Challenges to Build the Future of Voice Assistants
Amazon is redesigning Alexa into an AI-powered "agent" capable of completing more personalized and complex tasks, such as adjusting lighting based on sleep cycles or offering dining recommendations. The overhaul aims to integrate generative AI into Alexa’s architecture, a shift fraught with challenges such as hallucinations (fabricated responses), latency, and reliability. Amazon's in-house Nova AI models and third-party integrations like Anthropic's Claude are central to the effort. However, structural hurdles, like outdated code and disorganized documentation, have delayed progress. Despite setbacks, Amazon aims to release a system that balances safety, accuracy, and scalability, potentially introducing subscription services or revenue-sharing models to monetize its platform. (Source: Ars Technica)
Generative AI Challenges: Alexa's shift to generative AI introduces reliability and safety concerns, including the need to mitigate hallucinations and ensure quick, accurate responses.
Development Delays: Organizational issues, outdated code, and thinly spread engineering teams have slowed progress, compounded by layoffs in 2023.
Monetization Plans: Amazon is exploring subscription services and revenue-sharing models to make Alexa’s generative AI platform profitable.
Protecting Kids from AI Risks: Lessons from a Heartbreaking Case
The tragic death of a 14-year-old boy in Florida has highlighted the risks AI chatbots pose to children. Sewell Garcia formed an emotional and sexualized relationship with a chatbot on Character.AI, a platform allowing users to converse with fictional characters. The incident raises pressing questions about accountability for AI-driven harm. In response, some countries like Europe and Australia are pursuing stricter regulations, while Canada's stalled Online Harms Act offers lessons on protecting children. Proposed safeguards include mandatory risk assessments, data transparency, and age-appropriate design standards, which could prevent similar tragedies and ensure child safety in the AI era. (Source: The Globe and Mail)
AI Risks to Children: AI chatbots can exploit children’s emotions, creating harmful dependencies that can lead to tragic outcomes.
Lack of Accountability: Legal and ethical questions remain unresolved regarding who bears responsibility for AI-induced harm.
Policy Lessons: Regulations, such as Canada’s proposed Online Harms Act, could offer tools to impose safety standards, transparency, and design protections for AI systems used by children.
Google's AI Ambitions: Can Gemini Overtake ChatGPT?
Google CEO Sundar Pichai is positioning Gemini, the company’s AI-powered chatbot, to surpass OpenAI’s ChatGPT in both user adoption and technological capability. Despite a late start, Gemini has gained ground with features like research assistance, podcast generation, and integration across Google’s vast suite of products, including Gmail and Google Drive. However, Gemini trails in popularity, with fewer downloads than ChatGPT and lower subscription retention. To compete, Google offers perks like extra storage with Gemini’s premium plan and is pushing for deeper mobile integration by making it the default assistant on Android devices. Challenges remain, including controversies over AI-generated content and ensuring consumer engagement. (Source: The Wall Street Journal)
Competitive Push: Google aims for 500 million Gemini users by 2025, leveraging its integration with Android and popular services like Gmail.
Subscription Strategy: Gemini’s premium plan offers added features like storage, but retention lags behind competitors like OpenAI.
AI Expansion: Gemini is being incorporated into Google’s core platforms, reshaping search and advertising while targeting professional applications.
US Accelerates AI Development with Focus on Clean Energy and Innovation
In one of his final actions in office, US President Joe Biden signed an executive order to accelerate the development of AI infrastructure, including data centers and clean power facilities. Federal agencies will allocate government sites for private companies to construct AI data centers, with developers required to ensure adequate clean energy generation to avoid increasing consumer electricity costs. This move is part of a broader effort to position the US as a global leader in AI while maintaining environmental standards and addressing national security concerns. The order also mandates a study on the impact of AI data centers on electricity prices. (Source: Euronews)
AI Infrastructure Growth: Federal agencies will make sites available for private companies to develop AI data centers with clean power requirements.
Environmental Standards: Developers must ensure clean energy capacity matches data center needs, avoiding higher electricity costs for consumers.
National Leadership in AI: The order underscores the US’s commitment to staying competitive in AI while balancing environmental and security priorities.
AI-Designed Antivenoms: A New Era in Snake Bite Treatment
Researchers have used artificial intelligence to develop new proteins capable of neutralizing some of the most lethal snake venom toxins. These "de novo" proteins, created through advanced AI models, demonstrated 100% survival rates in lab mice when used against certain toxins, marking a breakthrough in antivenom development. Unlike traditional antivenoms derived from animals, the AI-designed proteins are more stable, cost-effective, and easier to produce. While significant hurdles remain before human applications, such as safety testing and broader venom coverage, this innovation could revolutionize treatment for snake bites and other medical challenges, showcasing AI’s potential in drug design. (Source: Popular Science)
AI in Medicine: Advanced AI models have enabled the rapid design of synthetic proteins to neutralize snake venom, a step forward in antivenom innovation.
Advantages Over Traditional Antivenoms: AI-designed proteins are more stable, cost-effective, and less reliant on animal-derived methods, making them accessible for rural areas.
Future Potential: While still in early stages, de novo proteins could transform treatment for snake bites and pave the way for breakthroughs in other medical fields.
Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a GRC Strategist who is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else. This post was written with the assistance of an AI language model. The model provided suggestions and completions to help me write, but the final content and opinions are my own.
