NTT Communications Confirms Cyberattack Exposing Data of 18,000 Corporate Customers
Japanese telecom giant NTT Communications reveals data breach affecting 18,000 corporate customers, with unknown number of individuals impacted
Elliot Kim
The release of DeepSeek's generative AI models has sent shockwaves through the tech community, leaving engineers and developers wondering how the company achieved such a breakthrough and how they can leverage the technology in their own stacks. The DeepSeek team built upon existing developments in the AI community, but took them to the next level, creating a model that rivals leading models like Meta's Llama 3.1, but at a fraction of the cost.
The significance of DeepSeek's achievement lies not only in its technical prowess but also in its open access approach. By releasing its work as open access technology, DeepSeek has opened the door for others to learn from and build upon its innovations, creating a more competitive market for large language models (LLMs) and related technologies.
So, how did DeepSeek achieve its breakthroughs? The company released two models: DeepSeek V3, a powerful foundational model comparable in scale to GPT-4, and DeepSeek R1, designed specifically for complex reasoning and based on the V3 foundation. The technical strategy behind each model is a testament to the team's innovative approach. DeepSeek V3 leveraged eight-bit precision matrix multiplication for faster operations, implemented custom logic to accumulate results with the correct precision, and utilized WGMMA parallel operators. The model also took multi-token prediction to the next level, inspired by Meta's French research team, and expertly used the concept of "common knowledge" to push the boundaries of Mixture-of-Experts (MoE) models.
DeepSeek R1, on the other hand, introduced a new approach to reasoning at scale, learning from a basic reward model, a first at this scale. The model's ability to realize on its own that spending more time thinking leads to better answers is a remarkable achievement. The incorporation of cold-start data from DeepSeek V3 also played a crucial role in making the model work.
The implications of DeepSeek's breakthrough are far-reaching. As Florian Douetteau, co-founder and CEO of Dataiku, notes, companies need to maintain an agnostic strategy with their AI partners to avoid vendor "lock-in" and ensure optionality in their AI journey. A multi-LLM infrastructure is essential to future-proofing LLM decisions and integrating new models as the market evolves.
Moreover, the rapid pace of innovation in the AI landscape demands rigorous testing, robust guardrails, and continuous monitoring to maintain control and governance. As the world of agentic AI continues to evolve, engineering teams must be prepared to adapt and innovate.
In conclusion, DeepSeek's breakthrough in generative AI marks a significant shift in the industry, offering a competitive alternative to leading models and paving the way for a more open and collaborative approach to AI development. As the technology continues to evolve, one thing is clear: companies that adopt an LLM-agnostic approach and prioritize control and governance will be best positioned to capitalize on the opportunities presented by innovations like DeepSeek.
Japanese telecom giant NTT Communications reveals data breach affecting 18,000 corporate customers, with unknown number of individuals impacted
Zimbabwe plans to pay $331 million to white farmers whose land was taken back in 2000, in a bid to restructure its massive debt and rejoin international markets.
Apple's generative AI platform arrives on Vision Pro, enhancing workflow and image generation capabilities, with a new iPhone app for seamless content management.
Copyright © 2024 Starfolk. All rights reserved.