China's DeepSeek has rocketed to the top of Apple’s App Store charts in the U.S., dethroning OpenAI’s ChatGPT as the most downloaded free app. The supposedly super-efficient, open-source software is raising some serious questions about the stock market valuations of key tech giants, including the chip maker Nvidia, whose stock got crushed Monday.
Designed for solving complex problems, DeepSeek’s new R1 reasoning model reportedly performs as well as OpenAI’s o1 on certain benchmarks. R1 was built on the V3 LLM, which the company claimed is on par with GPT-4o and Anthropic’s Claude 3.5 Sonnet, but cost less than $6 million to develop. By contrast, OpenAI CEO Sam Altman has said GPT-4 cost over $100 million to train.
See also: How AI Will Transform Insurance in 2025
The big question that's causing most of the concern is whether the entire industry has been wildly overspending and investing. DeepSeek is also raising profound questions about how China may have undercut America’s (and emerging U.K., Europe and likely other markets) most critical economic advantage on A.I. by making its technology free.
There's so much to unpack and some massive questions to answer about the true cost, model implications, trust, how DeepSeek can be harnessed from a business point of view, the censorship people are already reporting and on and on. Until we have more information, it will be very difficult to get to the bottom of any of these things.
While DeepSeek has reported some impressive results, we don't yet have much of a handle on its performance. These models often return different results on multiple uses, with pretty varying performance.
It is certainly way too early for any stock market calls. Let's hope the trading algorithms don't get carried away. Another "fake" market move is exactly what we don't need. Once we start getting more information and answers we can start to form more considered views.
For now, we need to keep exploring all of these models. For most of us harnessing multi-AI agent strategies, we need to look in-depth at everything in this space and ask more questions of our technology foundations, data models, security and so on.
These systems are, at the moment, hard to protect or control, which makes it harder to stop them from doing something wrong, or at least interrupting some output. By the way, these systems will potentially be operating at massive scale.
See also: Who's Getting Results From AI, and Why?
Existing regulatory frameworks even now make the accountability and responsibility pretty clear.
That's why insurers, for example, need to think super hard about designing AI systems differently, and designing their businesses around these technologies differently, as well.
They have huge societal implications. People, finances, vulnerability, risk understanding, health and so on are all potentially affected.
We need to be human-centered. We need to think really hard about how we build core systems and foundations that can harness the AI, keep humans in the loop and in control, allow multi-agent strategies to be realized, and make it easy to keep pace with the rapidly changing AI landscape.
The AI you are using and adopting today will become defunct in the next year. We know that processing speed and intelligence are set to rapidly change. And we know that countries are pouring billions into this technology.
Keeping pace will require huge amounts of core adaptability, in infrastructure like energy, as well as in the way businesses are built. This sort of required enterprise design change rarely exists in insurance.
There will be a lot more to follow on this as people begin to explore trust, usage and acceptance and really what's behind DeepSeek.
It has definitely put the cat among the pigeons.
