Yeah, MOE models have been around for a while, and how much this cost DeepSeek is unknown but surely is well above what they're divulging.

Yeah, I think rather than Deepseek slowing down investment it is going to accelerate it. It is obvious there is going to be an AI-race between different companies/organizations. 

The important thing here (as Yann LeCun noted) is that open source is competitive with proprietary. 

Just like open source compilers and programming languages didn't hinder the information-age era, but rather accelerate it so is it going to be true for AI. 

The field is broad and deep enough that all Chinese, American, and European (if Europe chooses to) tech organizations (won't say corporations because AI likely will be disruptive of the current iteration of capitalism) will be able to find their own niche. 

I know this is something unusual to say in our current era of nationalism and isolationism. 

The issue is a company managing to create an AI model which is just as good, or simply good enough, when compared to the competition, at a fraction of the cost and resources, so why are people like Sam Altman asking for 100s of billions of dollars? US companies have been leading the AI battle and that has been driving their valuations, now investors will wonder just how far these companies really are in the lead, with tens of billions being spent with zero profits in return for years on end.

Of course AI won't die but it shows a little fragility in the US companies hold on the AI market and this after all the song and dance about $500bn being invested into AI by US Companies whilst OpenAI loses $5bn+ per year. OpenAI is just one AI company though, but it is the most well known and Sam Altman comes across as a massive con artist and I do not believe OpenAI will be the one to make the breakthroughs and the problem is, a lot of companies are hedging a lot on OpenAI.

Nvidia is down 16% now, I'm sure it will rebound but there's fragility being shown for US companies.

Nvidia Stock Plunges 15% As NVDA Heads To Biggest Market Cap Loss Ever

This is basically what the Chinese team did if their claims are accurate, Lol. China also had US sanctions on them which was meant to weaken their ability to do this. It appears as though US companies are running extremely expensive models and at risk at being beaten out by more competitive inexpensive models. OpenAI, if these claims are accurate will need to sort themselves out and become far more cost efficient.

Broadcom Inc down 18%.

This is another example of how many in the U.S. are still stuck in the 1950s. They don't get that they have international competition. They think they still live in a world where Europe and Asia are still in ruins from WWII. That's how Japan was able to eat the lunch of U.S. industries in the 1970s and 1980s. At several points before the lost decade, Japan was on the cusp of becoming the largest econony on Earth, something that modern-era China has yet to accomplish. That's part of why Reagan made them sign the Plaza Accord.

@Ryuu96 I followed you on BlueSky.

Yeah it does feel like the US companies got caught with their pants down but we'll have to see how this develops, I'm not saying this will hurt AI, in fact it could be a good thing for AI, a cheaper access for more companies and cheaper access for users of AI. OpenAI currently charges $200 a month for their Pro Option on ChatGPT and they're still unprofitable! Lol. But it could be a bad thing for those in the lead who have been investing tens of billions into an unprofitable venture only to be undercut by a far cheaper and just as efficient rival.

All while China has restrictions too, Lol.

Are you the one I just followed back? Making sure.

None of the people who keep up with this technology are surprised that the cost for reasoning models is coming down. The surprise is mostly that it is coming from a Chinese company, as Chinese companies have been behind for a while now and they're under a chip embargo. These advancements happen in cycles and advancements are made across companies, not within a single one. 

Basically you have Google and Nvidia doing fundamental research. They're the ones really pushing the boundaries of what we know. 

Open AI/Anthropic/Meta/Mistral and now Deepseek are basically implementing this fundamental research into practical applications. This follows a cost-cycle, where if you want to be the first-to-release (as OpenAI tends to want to be) it costs a lot, but then as optimizations are discovered you can reduce the cost over time. So Open AI released o1-preview (an iteration of a model they first innovated in November 2023) in September and open-source has pretty much caught up with it in January at a cheaper price. That's not surprising. o3 will release in February, and you basically have the companies leap-frogging each-other until the next paradigm shift occurs (which seems to be Agents/Operators and/or an implementation of Google's Titans architecture.) Then the cycle repeats itself again until we get to the point of recursive self-improvement.

I do think there will be a bubble bust, just like the Dot-Com bubble or any other capitalist business cycle, but the market will exist nevertheless. 

Yes.

I wonder how Musk would see this painting.  I actually just saw it on FB and couldn't believe how horrible the picture is.  I remember a guy coming to our school to speak about his time in the concentration camps and the only thing that i remember (was 8 years old) was that he said that he had to kill another man and was crying as he told us.  Anyway, some tough jew should kick his ass.