I think this is just a stark example of how the AI sceptic crowd in the media believe things for political reasons. "We don't like big tech and big tech is making this thing, therefore it is bad and rubbish. A news story shows how much more potential this thing has - in a way that meaningfully undermines one of my criticisms - and somehow this confirms my prior beliefs".
I'm going to shamelessly steal and expand on a reddit comment that I saw earlier which I think had a good perspective on this situation:
"It's a bit like there's rumours of an AI gold rush and the big American companies have headed off on an expensively funded expedition to Alaska.
DeepSeek have come along and effectively said "Hey guys, I've borrowed some of your shovels and I've found just as much gold as you have digging in my back garden". "
Meanwhile, we're not sure that the gold is actually useful for anything yet - but we now know that the American companies have overpriced it by overbuilding prior to demand. It's a first adopter problem.
The big winners here aren't necessarily the AI sceptics, but they're certainly the AI Company Sceptics.
The AI sceptics will switch between being AI sceptics and AI company sceptics whenever it is convenient for them. This week they're being AI company sceptics.
Yeah I think this is bang on. I'm not a VC strategy expert, but my guess would be OpenAI's crazy valuations and tech growth relied on the idea of one of them "winning" and the moat being so big nobody could cross. DeepSeek shows that it doesn't matter how big the moat is: the drawbridge is down. It's a bit like if an Uber competitor back in the day turned around and said "we can do what Uber does but we pay our driver's a 30th of what they do", right? Uber would be pretty sad about that, but ride hailing would do quite well.
Shamelessly stealing the top comment from Hacker News, but if this new model is 20x more efficient than o1 at comparable performance, then we're probably soon about to see what 20x more compute will get you with GPT-5 (or whatever they call it).
If the answer is "not much", then that's when things will get ugly.
Oh my god! Everyone is freaking out about DeepSeek. It’s a deep fake hype and doesn’t even work well. It’s like everything else made in China, looks good on the outside but doesn’t actually work. China got what it wanted. Steal massive amounts of data and disrupt market… mission accomplished!!! They’ve already had to shut it down temporarily due to a quote un quote cyber attack.
I guess the question is that the amount of money being funnelled into AI companies in the west has been justified on the basis of "we need all this money to make it work", but if it's the case that the amount of money being poured in is genuinely humongously more than is needed, then the extra compute power won't just be poured into other uses, but rather people will see it as so incredibly wasteful that they'll be more skeptical of AI investment altogether, at least AI investment without more concrete results
I think this is just a stark example of how the AI sceptic crowd in the media believe things for political reasons. "We don't like big tech and big tech is making this thing, therefore it is bad and rubbish. A news story shows how much more potential this thing has - in a way that meaningfully undermines one of my criticisms - and somehow this confirms my prior beliefs".
I'm going to shamelessly steal and expand on a reddit comment that I saw earlier which I think had a good perspective on this situation:
"It's a bit like there's rumours of an AI gold rush and the big American companies have headed off on an expensively funded expedition to Alaska.
DeepSeek have come along and effectively said "Hey guys, I've borrowed some of your shovels and I've found just as much gold as you have digging in my back garden". "
Meanwhile, we're not sure that the gold is actually useful for anything yet - but we now know that the American companies have overpriced it by overbuilding prior to demand. It's a first adopter problem.
The big winners here aren't necessarily the AI sceptics, but they're certainly the AI Company Sceptics.
The AI sceptics will switch between being AI sceptics and AI company sceptics whenever it is convenient for them. This week they're being AI company sceptics.
Yep, and if the climate argument disappears they'll just seamlessly make copyright their main gripe.
I’ve already seen endless copyright gripes
Yeah I think this is bang on. I'm not a VC strategy expert, but my guess would be OpenAI's crazy valuations and tech growth relied on the idea of one of them "winning" and the moat being so big nobody could cross. DeepSeek shows that it doesn't matter how big the moat is: the drawbridge is down. It's a bit like if an Uber competitor back in the day turned around and said "we can do what Uber does but we pay our driver's a 30th of what they do", right? Uber would be pretty sad about that, but ride hailing would do quite well.
Shamelessly stealing the top comment from Hacker News, but if this new model is 20x more efficient than o1 at comparable performance, then we're probably soon about to see what 20x more compute will get you with GPT-5 (or whatever they call it).
If the answer is "not much", then that's when things will get ugly.
It's besides the point, but I cannot accept ‘compute’ as a noun. It does not compute.
Oh my god! Everyone is freaking out about DeepSeek. It’s a deep fake hype and doesn’t even work well. It’s like everything else made in China, looks good on the outside but doesn’t actually work. China got what it wanted. Steal massive amounts of data and disrupt market… mission accomplished!!! They’ve already had to shut it down temporarily due to a quote un quote cyber attack.
I guess the question is that the amount of money being funnelled into AI companies in the west has been justified on the basis of "we need all this money to make it work", but if it's the case that the amount of money being poured in is genuinely humongously more than is needed, then the extra compute power won't just be poured into other uses, but rather people will see it as so incredibly wasteful that they'll be more skeptical of AI investment altogether, at least AI investment without more concrete results