Deepseek from Vajirao & Reddy Institute

DEEPSEEK

• DeepSeek is an AI startup based in Hangzhou, China, that has recently gained global attention for its innovative and low-cost AI models.
• The company introduced its AI models—DeepSeek-V3 and DeepSeek-R1 (a reasoning model)—which are seen as potential competitors to OpenAI's advanced models like GPT-4.
• What sets DeepSeek apart is its ability to achieve similar performance to OpenAI's models at a fraction of the cost.

• Founding and Focus:
  • DeepSeek is a startup from Hangzhou, China, which has launched a series of AI models that excel in tasks such as math, coding, and reasoning.
  • Its models are powered by a low-cost Large Language Model (LLM) infrastructure, which makes them more affordable than many global counterparts.
• Comparative Edge Over Global LLMs:
  • DeepSeek’s models are designed to be far more cost-effective than competitors like OpenAI’s GPT-4.
  • Training Cost Comparison:
    • DeepSeek: $6 million
    • Global LLMs (e.g., GPT-4 by OpenAI): ~$100 million
  • This significant cost difference is primarily due to DeepSeek’s use of older-generation hardware (NVIDIA H800 chips) compared to the more advanced GPUs used in OpenAI’s models.
• Cost and Accessibility:
  • Subscription Cost:
    • DeepSeek: $0.50 per month
    • OpenAI's ChatGPT: $20 per month
  • The affordability of DeepSeek’s services allows for broader accessibility, especially in regions with budget constraints.
• Training and Performance:
  • Training Approach: DeepSeek uses reinforcement learning to enable its models to self-improve and adapt, which contrasts with the supervised learning model used by OpenAI.
  • Performance: DeepSeek’s models are comparable to OpenAI's o1 model in many performance metrics, though they are not yet as advanced as the o3
  • Scalability: DeepSeek focuses on creating smaller, faster models (SLMs), which are more resource-efficient and scalable.

• DeepSeek Coder: A model designed for coding-related tasks.
• DeepSeek LLM: A 67-billion-parameter model intended to compete with other large language models.
• DeepSeek-V2: A cost-effective model with strong performance in a variety of tasks.
• DeepSeek-Coder-V2: A 236-billion-parameter model designed for complex coding challenges.
• DeepSeek-V3: A 671-billion-parameter model capable of coding, translation, and generating essays/emails.
• DeepSeek-R1: A reasoning model aimed at challenging OpenAI’s o1 model.
• DeepSeek-R1-Distill: A fine-tuned version of DeepSeek-R1, based on synthetic data generated by R1.

• Censorship and Bias:
  • DeepSeek adheres to China's strict digital content regulations, which means it avoids providing direct answers on sensitive political topics.
  • This adherence to government censorship raises concerns about biases in the AI’s output.
  • There are fears that DeepSeek's models might carry a pro-China bias due to government influence over the technology.
• Security Risks:
  • Experts have expressed concerns over potential security risks, particularly related to data privacy and the ethical use of AI.
  • Given DeepSeek’s origin in China, these concerns are amplified due to the broader context of global geopolitical tensions.

• A Large Language Model (LLM) is a type of artificial intelligence model that is trained on massive datasets containing text data.
• LLMs use deep learning techniques, particularly neural networks, to understand, generate, and process human language.
• These models have billions (or even trillions) of parameters, which allow them to perform a wide range of language-related tasks, including text generation, translation, question answering, and more.
• Examples: OpenAI’s GPT-4, DeepSeek’s models, and Google's PaLM are examples of LLMs that have revolutionized natural language processing (NLP) tasks.

• Sputnik Moment: The launch of DeepSeek has been compared to the impact of the Soviet Union's Sputnik launch in the 1950s, marking a shift in the technological competition between global powers, particularly between the US and China.
• Market Disruption: The introduction of DeepSeek’s AI models caused a significant drop of $600 billion in the market value of Nvidia, a leading manufacturer of AI chips.
• This highlights the growing importance of AI in shaping the tech market and how companies like DeepSeek are challenging established industry giants.
• Policy Implications: DeepSeek’s rapid advancements could trigger further restrictions on AI and semiconductor technology exports from the US to China, heightening the ongoing rivalry between the two nations.