Jan
28
DeepSeek , Donald Trump and me!
Filed Under Uncategorized | Leave a Comment
It has been a difficult few days. Midway through last week I was feeling a rising emotion of joyful exhilaration. Despite the sanctions, despite the US technological lead, despite the TikTok war, China was fighting back. and demonstrating a technological resilience which I had discussed last month in an Outsell interview with a Chinese AI researcher in Beijing.Surely just what was needed to counterbalance the sheer arrogance of the praetorian guard of tech billionaires flanking the US President?
The weekend was spent watching the feedback. As Nvidia plunged, my thoughts turned to events switch ignite recessions, and the fact that all of this was likely to impact investment, markets and pension funds – including mine. Elation turned to depression. I postponed the grocery delivery order by a day to wait out events. With this new week dawning, Donald Trump interrupted his 24 hour newscast to say that DeepSeek was “a good thing“. What is this? Am I agreeing with Donald Trump? My weekend sense of depression rapidly turned to despair as I scrabbled two read all I can find in the past 24 hours in order to try to restore my mental stability. What followers is the result!
DeepSeek does not seem to be a technological breakthrough. It appears to be an extremely clever piece of reverse engineering. It is the type of process which happens throughout all phases of technological change. And it appears to have happened in China before the Biden ban on high-tech chips really began to bite. High Flyer, the incubator parent of DeepSeek, is an investment company and would not have been one of the places where observers would have looked for a high-tech response to Silicon Valley. This part accounts for the shock. But we might have expected someone to use the advances that Silicon Valley has made, at great expense, and reproces them to get more cheaply to the same conclusion. Would there have been a Japanese car industry without Dearborn?
But surely the important thing is that DeepSeek takes us to where ChatGPT is, but in an open source context. This is vitally important in terms of the widespread use of the release.. This alsobegs another question: what is the business model of DeepSeek? As far as I can see, nobody knows. Yet since nobody does very much unintentionally in China, I must presume that somebody somewhere does know. Apart from Chinese altruism (yes, you did just read those two words in the same sentence) there may have been a desire to spike US confidence at a moment of chronic rhetorical overstatement. Or, if you believe in the ability of TikTok to collect dangerous data on behalf of the Chinese state from US citizens, what would be the effect of prompting the greatest numerical download of software to US computers from a Chinese source ever?
Leaving aside the unanswered questions I’ve also found myself in the last 24 hours trying to create a balance sheet. The idea that a really effective open weights AI environment is available to the world’s scientist very cheaply must be a huge plus. I grant that it may not be found to be as effective as claimed, or as cheapto operate as claimed. But for many of us, there is a huge plus in deploying niche based sectorized AI using small learning models to automate definable process activity. For reasons of economic growth and productivity (ask the UK government) we want applications rolled out faster and cheaper, so that we can detect cancer before the human eye can see it on a scan, engineer our tax filings, sort out logistics and transport problems and solve the problems of a society where expectations surge while solutions creak.
For those who have been spending billions, including the US government, in funding the undeniably expensive business of leading edge AI development, this is a sobering moment. It appears that the lead time before the rest of the world re-engineers your solutions and catches up is about 2- 3 years. The days when you had a decade to exploit a technology breakthrough are over. Just as the speed of technology development is growing, so the speed of technology catch up is quickening.
The next bit is really scary. As Mr Trump calls for a speeding up of competitive responses, and removes entirely the very modest Biden administration safety rules on AI releases, there must surely be real fears that either the Chinese or the Americans will release prematurely a piece of unvalidated intelligent code which has a disastrous effect upon the world’s systems and services. Then we can really talk about recession. This is not about the
deregulation: it is about the basic safety rules that secure society against self harm. A world that keeps on talking about national sovereignty as if we were in the 1890s ignores at its peril the digital revolution which since the late 1990s has interconnected all of us in a network configuration whose implications it seems we keep on trying to evade or avoid.
So for the ever anxious futurist – people like me – there is some relief (I do not in fact agree with Donald Trump after all) and the sense that we must work to restore the focus once again. The arrival of DeepSeek is momentous, especially if it allows us to cheaper , smallscale research and applications enabling real advances that change peoples lives in a day-to-day sense. But perhaps it over shadows what should have been the subject we were all thinking about this week. The launch of Open AI Operator is the opening salvo of a new development war: it is a push into the agented world that many of us feel will be a hallmark of the next five years. Indeed, many of us will never touch AI in the sense of becoming users of DeepSeek or anything else. Perhaps for most of us change will come through the agents we use and the interactions that they make, and the intelligence that they display. If you are, like me, worried about AGI and unsafe systems and in your 80th year, then a stiff whiskey helps. If you want to think about the impact on society of AI in the next five years, then studying developments like Operator and Rabbit-R1 maybe more to the point.