the Deep seek Rabbit Hole just keeps getting deeper and deeper after being deeply wounded by Deep seek last week open AI is now reportedly accusing them of Ip theft and to make matters even worse a second Chinese model just hit the timeline if you're just waking up from a coma the legend goes like this a Chinese hedge fund built a state-of-the-art reasoning model that surpassed open A1 and only spent $5.5 million to train it then gave the world a 100% discount code to use it that was devastating to Big Tech and especially open AI who
have been gradually trying to convince people that AI is hard and we need things like $500 billion Stargate data centers now in order to get the AI hype train back on track the new White House aiar and PayPal Mafia member David Sachs just went on the news and said they have substantial evidence that deep seek stole open ai's outputs to fine-tune their models a technique known as distillation that is strictly forbidden in their terms of service the irony here is not only palpable one might describe it as artificial super irony you might remember how open
aai vacuumed up the entire internet and all its copyrighted materials without asking anybody for permission Elon them of scraping Twitter George RR Martin and a bunch of other authors are suing among many other lawsuits around the world but I need to let you in on a tech bro founder secret we do Shady stuff and then ask for forgiveness later because once a company reaches critical mass there's no stopping it Uber and Airbnb are prime examples and thus far open aai has mostly prevailed in their copyright infringement legal battles in today's video we'll take a closer
look at the technical details of deep seek how it bypassed Cuda and try to find out if it actually ripped off open AI it is January 29th 2025 and you're watching the code report my conspiracy theory is that open AI has been deep seek all along and this was just the most genius marketing trick of all time to assert their dominance I wouldn't put it past Chief persuasion officer Sam ultman he's a he's a just a Kong man and he lies to everyone but now open Ai and Microsoft are accusing deep seek of distillation where
you take one big expensive model like 01 and use its outputs to transfer knowledge to a smaller model thus far they haven't provided any hard evidence but there are screenshots like this going around the internet where deep seek provides a response that should only come from chat GPT that's not a Smoking Gun though because this type of content is all over the internet now is so deep SE could have learned it organically however Microsoft which provides the servers for much of open ai's compute is said they observe someone in China extracting large volumes of data
from the open AI API and they believe these accounts may be linked to deep seek in other words deep seek is basically Robin Hood stealing from the rich to give to the poor the distillation generally provides better results compared to reinforcement learning where you actually feed the model new data with a reward function distillation is not controversial and deep has models distilled from llama and quen in fact you can even distill open AI models as long as you don't use the API to build a rival model and that appears to be the root of open
ai's beef and get this Alibaba just released quen 2.5 Max just minutes ago and although it's not a reasoning model it's an open model that beats deepseeker Claude and GPT 40 on these benchmarks and not only that but yet another Chinese model called Kim 1.5 just came out and apparently beats open AI 01 we're now in a China versus China AI race with the United States falling behind meanwhile Europe is focused on other technological innovations like bottle caps you can't take off many people have complained that deep seek is highly censored but it's relatively easy
to jailbreak if you're a senior prompt engineer but speaking of irony last year mid Journey accused stability AI of image theft but none of that matters now because deeps also just released the Jan series models which do diffusion based image generation and while the quality is not as good as stable diffusion or mid Journey it's yet another open source model you can use commercially and that's good news for Humanity but another interesting detail about deep sea is that it achieved 10x better efficiency than other models in part by not using Cuda nvidia's proprietary platform for
running code on a GPU instead they used Nvidia parallel thread execution directly which conceptually would be similar to building a website with assembly code and is just another example of how crack these deep seek Engineers really are now another major criticism of deep seek is that when you use it on the web all your prompts data and keystrokes go to China if you care about privacy though you shouldn't be using the internet anyway instead you should use it locally like I did in this video on my second Channel but the most important Trend here is
that open source is winning that means if you're a developer now is the time to start building products people love and you can do that with a truly awesome open source tool called post hog the sponsor of today's video it's like a Swiss army knife to analyze test observe and deploy better features its product Analytics tool can help you understand your customers and build funnels its web analytics can replace Google analytics and session replay will help you understand how users actually interact with your app not to mention feature flags and UI experiments like AB test
just to name a few of the features but most importantly it's easy to implement thanks to sdks for web mobile and serers side apps with excellent docs designed for developers not only is it open- source and self- hostable but also has a fully managed no card required free plan give post hog a try with the link below this has been the code report thanks for watching and I will see you in the next one
