DeepSeek: Regarding DeepSeek’s claims of cost-effectiveness and open-source approach. Even if DeepSeek doesn't immediately change your AI roadmap, does it spark any curiosity about exploring new possibilities or use cases?
Sort by:
I agree with those sentiments. DeepSeek seems like a natural evolution, and I appreciate its emergence despite the controversy surrounding its origins and funding. It serves as a challenge to the core players in the field, helping maintain a competitive edge. Language models are evolving rapidly, and I believe DeepSeek's impact will be more significant in driving competitive behavior forward in the long term. While it has made advances in compute efficiency and cost reduction, for many AI consumers, including us at Bluestem, the real cost lies in the vector databases on the back end. While it's beneficial for larger players to learn from and reduce operating costs, the impact on consumers like us may be limited from a cost perspective.
I'll echo what Sean said. In the federal space, we are cautious of unknowns due to regulatory requirements. From a regulatory standpoint, I'm curious but not willing to take too many risks. However, as a small woman-owned business, I hope it helps reduce costs. Balancing compliance and cost is challenging, but it's something we have to navigate carefully.
The cost-effectiveness of new AI technologies could indeed make them more accessible to smaller companies. Their agility allows them to capitalize on advancements more rapidly, potentially outpacing larger competitors.
Smaller companies, despite budget constraints, might actually be better positioned to leverage AI quickly. They aren't bogged down by legacy systems and can be more agile. While security is critical, the nimbleness of smaller companies allows them to take calculated risks and potentially gain a competitive edge. They can explore new verticals and capture market share more swiftly than larger enterprises.
Security remains a paramount concern. Large companies might have the resources to experiment with new technologies, but for mid-sized and smaller enterprises, the costs can be prohibitive. Without leading with security and policy, there's a risk of unintended consequences. Companies need clear policies on AI usage, especially regarding sensitive information. Establishing governance frameworks, as suggested by Gartner, is essential before diving into new technologies.
I read an article suggesting that DeepSeek wasn't entirely transparent about its use of fewer CPUs, omitting some pre-training details. This reminds me of how countries once circumvented supercomputer export restrictions by exploring parallel computing. It suggests that there are alternative paths in AI development, encouraging innovation and diverse approaches.
I'm not sure if it sparks new curiosity, but about a year and a half ago, I attended a major vendor's AI roadshow where they discussed various language models. That's when I first became aware of small language models. I feel like DeepSeek, while it has made waves, is part of the natural evolution of this field. It starts in a certain way, then becomes more efficient, optimized, and expands its capabilities. However, there are many unknowns about DeepSeek right now. Questions about how it is trained remain, and despite its open-source approach, much is not yet transparent.