DeepSeek Is About to SHOCK THE WORLD With R2 That’s 40X More Efficient Than OpenAI's AI

Deepseek is racing to drop its next AI model ahead of schedule, undercutting OpenAI by up to 40 times in cost and shaking up the OpenTire industry. Meanwhile, Alibaba's new video AI is already outperforming OpenAI's Sora, and Western governments are starting to panic. And as OpenAI rolls out new research tools and voice features, one big question looms. Just how persuasive should AI really be? Alright, first off, Deepseek has been making headlines in a major way. If you remember, they launched their R1 model in January and it basically caught everyone off guard. R1 was touted as a powerful AI reasoning model, and it was training at a fraction of the cost that bigger companies like OpenAI reportedly invest in their own models. Some people, including Google, were skeptical of those claims. Google even called Deepseek's statements exaggerated, and OpenAI suspected that DeepSeek might have used distillation from ChatGPT's infrastructure. But. But either way, the model was out there, and big names like Microsoft added R1 to Azure, AI Foundry and GitHub, and Amazon Web Services also featured it in their model catalog. Now, the big story is that Deepseek wants to release their successor R2 earlier than they initially planned. They had said R2 might launch in early May, but new reports suggest they're trying to get it out even sooner. So sadly, we don't have an exact date, but the rumor is that it could be anytime before May. The upcoming R2 is supposed to have improved coding capabilities and be able to reason more effectively in languages beyond English. That's a huge deal because if you think about it, a lot of advanced language models tend to revolve around English, so having robust multi bangit support could position Deepseek as a serious global contender. Why is Deepseek pushing this timeline? Well, GPT 4.5 is still weeks away and GPT 5 might not arrive for months. So if R2 hits the market soon, Deepseek could once again shake things up in the AI ecosystem. And they've already proven that they can undercut OpenAI by a wide margin in terms of pricing. According to analysts at Bernstein, Deepseek's pricing can be 20 to 40 times cheaper than what OpenAI charges for comparable performance. That cost saving aspect has drawn in not just small enterprises, but also major players wanting to integrate R1 into their offerings. Now, to really understand what makes Deepseek tick, you need to know a bit about its founder, Liang Wenfeng. He's been described as super introverted and low key but he became a billionaire thanks to his quantitative hedge fund High Flyer. The vibe is that he runs Deep Seek more like a research lab than a classic for profit startup. He even pays employees top tier salaries. Some senior Data scientists earn 1.5 million yuan annually, which when other rival quant funds usually cap around 800,000 yuan. He is also known for having a flatter corporate structure which is pretty different from the typical Chinese tech giant model with 9am to 9pm six days a week. Instead, people report working normal eight hour days in a pretty collaborative hands on environment. High Flyer, the hedge fund behind all this, actually poured a ton of money into AI research way before R1 made headlines. They spent around one and a half billion yuan on two supercomputing AI clusters in 2020 and 2021. One of these clusters, Fireflyer 2, consists of about 10,000 Nvidia A100 chips. This was before the US banned the export of those chips to China. So by the time that ban rolled around, High Flyer was already set. That gave them a big advantage. The key to Deepseek's cost effectiveness is its use of techniques like mixture of experts and multi head latent attention. In moe, you basically divide the model into specialized expert components so it doesn't have to tap the entire model for every single query. Meanwhile, MLA means the model can process different parts of an input simultaneously, picking out the most important details more efficiently. As a result, Deepseek claims it can achieve performance on par with bigger, more expensive models without breaking the bank. Chinese authorities, interestingly, are fully behind Deepseek. We're seeing municipal governments, energy companies and big corporations like Lenovo, Baidu and Tencent all integrate DeepSeek into their products. The government is even telling Deepseek to keep a low profile in international media. Meanwhile, some western governments like South Korea and Italy have restricted or removed Deep Deep Seek based apps over privacy concerns. There are also broader fears that advanced AI models could be used for social engineering or misinformation campaigns. So it's not surprising that scrutiny is intensifying in certain regions. But it's not just Deep Seek making moves. Alibaba recently announced their Open Source Video foundation model WAN 2.1, which is reportedly outperforming another OpenAI model called Sora on certain benchmarks. Alibaba's new offering includes multiple sub models optimized for text to video, image to video, video editing, text to image, and even video to audio. They have WAN2.1i to V14B and WAN 2.1. T2 V14B, which can both generate videos at 480p and 720p plus a smaller T2 V1.3B model that can run on consumer grade GPUs like in RTX 4090. According to Alibaba, WAN 2.1 can handle complicated motions and realistic physics simulations, and it's posted some excellent metrics on the vbench leaderboard. Part of the secret sauce is a novel 3D causal vae architecture with a feature cache mechanism for speed, plus a flow matching framework within a diffusion transformer structure. In short, they threw a lot of advanced techniques into the pipeline to training on about 1.5 billion videos and 10 billion images. That's huge. On the OpenAI front, there have been a couple of big stories. First, they expanded access to their Deep Research tool. Originally it was only for Pro subscribers at $200 a month, and now it's open to all paid ChatGPT users plus Teamadget Enterprise. If you have a plus plan, you'll get 10 deep research queries each month. Pro users get a bump from 100 queries up to 120. In case you're not familiar with it, Deep Research is basically an in depth tool for generating detailed, resource intensive reports, sometimes taking five to 30 minutes to compile a single query. The reasoning is a bit slower but can be more thorough, and now it also incorporates images in citations for a deeper dive. However, free ChatGPT users still don't get Deep Research because apparently it's super resource heavy. Another OpenAI update revolves around the topic of persuasion risks. In their recent whitepaper, they discussed how the Deep Research model might be used to persuade or influence people's beliefs. Because of that, they're holding off on launching an API version of the Deep Research model. They're basically reevaluating and how to test for real world persuasion, including how AI could produce extremely personalized content that sways opinions on a massive scale. They also ran tests like seeing if the Deep Research model could persuade GPT4O to pay money or reveal a code word. It actually outperformed some older OpenAI models at that, though it wasn't quite as persuasive at the code word scenario. Even so, OpenAI wants to tread carefully before making it widely available to outside developers. Finally, there's one more tidbit from OpenAI that's pretty cool. They're rolling out a preview of ChatGPT's advanced voice mode to free users of the platform. This feature had been exclusive to plus subscribers for a while, using a variant of the GPT4O mini model for speech. The advanced voice mode allows you to talk to ChatGPT out loud rather than typing all your questions. It can handle more natural real time conversations and even letting you interrupt mid responses to ask clarifying questions or pivot the topic. That said, there are daily usage limits and they can fluctuate, but it's still neat that more people get to test it out without paying for a premium subscription. If you want to try it, just tap the voice icon in the ChatGPT app, grant microphone access, and you're off to the races. Interestingly, Apple has also gotten involved in some sort of collaboration with OpenAI. Rumors are floating about an Apple OpenAI ChatGPT deal that might revolve around making AI voice interactions even more seamless on Apple devices. We'll have to wait for official confirmation, but it seems like there's something brewing there too. So prices are dropping, competition is heating up, and the big players are making moves that could reshape the entire landscape. Let me know in the comments. Do you think Deep Seek has what it takes to challenge OpenAI, or will the Western giants push back even harder? If you enjoyed this breakdown, drop a Like subscribe for more deep dives. Thanks for watching and I'll see you in the next one.