In all the AI innovation hype of the last two years, it’s been mostly American companies in the thick of it. The focus has been on ChatGPT and Google’s Gemini and their efforts to make generative AI mainstream. Then, seemingly from nowhere (although those following opensource LLM’s weren’t as surprised) China launched Deep Seek.
It’s competitive offering has the tech industry buzzing, especially with the claim of using older micro-processors and operating on a fraction of the cost. What’s really going on with Deep Seek? It’s is a contender to take the lead in the AI race and what is the potential impact on global AI innovation and development?
Some will say that competition, in any form, has the potential to advance innovation and change approaches to learning. What’s significant about Deep Seek is the transparency in its learning models. Whereas most US based companies developing AI have kept their LLMs proprietary, Deep Seek’s open-source approach provides a high level of transparency as to how decision making is made within the model.
Academia seems to be really excited by this, as it provides an opportunity for community learning and input as a way to make AI learning models more efficient. Specifically, efficiency isn’t just about getting to a correct answer faster, but also about becoming more accurate and relevant by being able to embed common sense. Something that human’s programming LLMs often overlook and only realize when AI starts spouting ridiculous answers.
Recently, we talked about the billions being promised to build bigger better data centers for AI, under the presumption that this is necessary for innovation and advancement. Traditionally LLM’s rely on massive datasets, with the mindset that better decision making requires even more data. Deep Seek has shown that by generating more efficient algorithms, the requirement for large scale and expensive infrastructure can be done away with. For the companies trying to justify the billions they’re asking for in investment, this does throw a spanner in the works.
AI also brings with it an energy heavy burden, and it’s this demand that has environmentalists being cautious about further AI data center development. But if AI can become more efficient, if algorithms no longer require such massive computing ability, then there is a chance that AI could be developed more sustainably. Whether they will, is another debate.
Other factors to consider are how AI is being integrated with automation. Already Chinese manufacturing brands have announced they’ll be using Deep Seek to refine their self-driving vehicles. One of the reasons that there’s wariness about self-driving cars is that people don’t trust the car to make the right decision. But if there’s greater transparency to that decision making process, as is the case with Deep Seek, then there’s potential for that perception to change.
Of course, we can’t talk about AI development without touching on privacy and security concerns. This is already a grey area, where many are questioning the safety of personal and company data once it’s accessed by AI. Arguably there’s currently insufficient governance and accountability. Now there are global players in the game, how is global governance to be defined and who will hold companies and governments to account? Italy and South Korea have already banned Deep Seek, will other countries follow?
