5
How DeepSeek trained its model
9d • macbudkowski.eth •

CMC is saying Deepseek is responsible for the general bleed in crypto today. Could potentially go down more.
Nvidia and tech that invested billions in AI went down, and crypto followed. It's an interesting question if this news isn't actually long-term bullish for AI because it can become more omnipresent long term, kind of like scaling Ethereum reduces the fees, but is long-term bullish.
> CMC is saying Deepseek is responsible for the general bleed in crypto today. Could potentially go down more. Why exactly?
Tech stocks are down and crypto is correlated.
It doesn't make too much sense for me the market is reacting this way. Why is Nvidia down because a breakthrough model in AI? Don't you need GPUs for any kind of AI work? And it's not like China wasn't a big influencer on AI models, with the video models they are much more advanced than any US based competitors. I think this is just the market correcting itself a bit before continuing, or before something truly major happens to shift the sentiment...
> It doesn't make too much sense for me the market is reacting this way. It may make economic sense if you consider the effects that the cost optimizations may have on the value of NVIDIA chips > "We still don't know the details and nothing has been 100% confirmed in regards to the claims, but if there truly has been a breakthrough in the cost to train models from $100 million+ to this alleged $6 million number this is actually very positive for productivity and AI end users as cost is obviously much lower meaning lower cost of access," Jon Withaar, a senior portfolio manager at Pictet Asset Management, said. (https://news.kiwistand.com/stories/DeepSeeks-Sputnik-moment-sparks-rout-in-AI-linked-stocks?index=0x6797ac83457f23b497831a4e027eaaf2ba9cad234f2aea2e974f6d0d3051a5c2c60ca081) If you can now train the same model as gpto1 with a tenth of the chips and you don't even need the latest generation, it means that the calculations for how many chips NVIDIA can sell in the short term has to be adjusted. Remember, the price to sales multiple of NVIDIA is 56x. Even Coinbase is evaluated more conservatively at a 52x multiple, and Coinbase is a really risky business. I actually think this fact that you can get to the same results as OpenAI with less chips also fundamentally collapses some AI chip investor's thesis, because they had fundamentally believed that it was chips which are the scaling bottleneck, and so that justified paying high multiples as they hadn't believed in software optimizations. Now it turns out that you just need a bunch of smart people to come up with more research and then you won't need more chips. All this said, the claims are from China so... we'll see