Loading summary
Tech News Host
I've been swimming in tech news all morning and I would say come on in, the water's fine, but it's not here. Let me dry off. While Chinese AI company Deepseek didn't fully burst the AI bubble, it did send Nvidia stock into a tailspin this week as investors questioned the need for Nvidia GPUs when Deepseek's models are just so dang efficient. But maybe part of the reason they're so efficient is because Deepseek didn't spend any time developing jailbreak defenses. AI security researchers from the University of Pennsylvania and Cisco found that Deepseek's reasoning model R1 completely failed to protect itself against a barrage of 50 popular chatbot jailbreak techniques, which I'll admit sounds very disorienting. Deepseek also apparently didn't devote any time to protecting their internal database of user chat histories and API keys, which was found by security researchers from Wiz just sitting there without even a password to protect it. Between that and the whole sending data to China thing, multiple countries are banning, restricting or investigating the use of Deep Seq models, including Italy, Taiwan, South Korea, France and even Belgium, who's usually pretty chill. But Deep Seq hype remains high and OpenAI had to do something. So they launched their O3 mini reasoning model, which surpasses O1, particularly in science, math and coding performance, and also doesn't immediately tell you how to make cocaine and explosives when you ask nicely, or worse, explosive cocaine. Limited use of O3 mini is free through ChatGPT, but paid users can use it more with $200 ChatGPT Pro subscribers getting unlimited use. Meanwhile, Microsoft now offers OpenAI's O1 model through the Think Deeper feature in its Copilot Assistant for free. Think. And despite saying it's investigating whether deepseek improperly used OpenAI models to train their own, Microsoft is also hosting DeepSeek R1 on its cloud services. But you know what they say, Keep your friends close and your enemies safely contained in a server farm. Nvidia wasn't kidding when it warned about stock shortages for the RTX 50 Series cards when they launched yesterday. Inventory for the RTX 5090 and 5080 sold out in minutes at retailers across the US, the UK and probably elsewhere. Sorry I didn't check every country. In some European countries, it almost seemed like stock was never there in the first place, thanks to the leak of a distributor order link. And if you're thinking, wow, people are more passionate about top tier PC gaming than ever oh my sweet summer child. This was reportedly due to incredibly low stock numbers and determined scalpers, with RTX 5090s being listed for as high as 9000 USD on eBay. YouTuber Jay Z FittySent tweeted yesterday that he spotted a 5090 he had just signed at a Micro center launch event that morning, being listed on Facebook Marketplace for $5,800. Now, all that despite the RTX 5080 representing objectively bad value compared to previous 80 series cards, as laid out so clearly in this glorious graph made by Paul's hardware for his recent video. Paul really knocked it out of the park with his graph. It's a good graph, Paul. But hey, it's not all bad news. Nvidia also, surprise, dropped a new feature in its GPU driver called Smooth Motion, a spatial frame interpolation feature which can bring less powerful frame generation to almost any game. Kind of like AMD's fluid motion frames, which would be cool if Nvidia didn't say that. For now, it's exclusive to the RTX 50 series, which is, you know, if you didn't catch it earlier. So out of stock stores are sending people home with paper vouchers just to feel like they got something. You ask me. Instead, they should pick up something from our sponsor. Oh, Snap. The company that knows phone accessories usually suck, so they designed one that doesn't. The Snap grip is crazy thin at 2 1/2 millimeters. That's smaller than almost any camera bump, so it won't get caught on your pockets. Hold it with two fingers or just one, I won't judge. It's also a variable kickstand and it's magnetic, so you can just use it with MagSafe accessories. Or just stick your phone to the front of your fridge because you can come on. And they're running a Valentine's Day promo right now. Buy one for yourself, get 50% off one for your partner@osnap.com TechLinked or by using the link in the description. I mean, it's the least you can do. I dried myself off, but I still feel like a sticky residue. Oh, it's cause there's still quick bits. Intel released their company financial results yesterday, and you can find the link in our news sources below. But I wouldn't, unless you just kind of wanna feel sad. TeamBlue says Panther Lake laptop CPUs are coming later in 2025, but Nova Lake, the successor to the truly depressing Arrow Lake desktop chips, aren't expected until 2026. Intel also announced that rather than competing with Nvidia's AI accelerators, they're canceling their Falcon Shores AI GPU for now, giving off some real Eeyore energy. It probably wouldn't have been any good, but hey, it's not like intel is done yet. Microsoft just announced their existing Surface Pro and Surface Laptop for businesses can be configured with Intel Lunar Lake processors instead of Qualcomm Snapdragon X chips. They'll just cost around $400 extra for the privilege of feeling like you've been Thanos snapped, but you're disintegrating extra slowly. Apple has filed a motion to delay proceedings in Google's antitrust trial after a judge denied the company's request to intervene in the case. It's a heartwarming story of massive tech giants sticking up for each other because one wants to continue getting paid $20 billion a year to have Google be the default search engine in Safari. Wow. I mean, without that money, Apple might have to cancel more projects like the Mac Connected AR glasses, which are no longer in development, according to master of Apple secrets Neil Germstrong. By which I mean Mark Gurman. Don't worry, Meta's Mark Zuckerberg told investors he plans on selling billions of AR glasses, so there'll be lots to go around. We'll just have little pairs. We just slap on babies right out of the womb. And you know what? Let's have some positive stories to round off this episode. After launching Spider Man 2 on Steam this week, Sony announced the game won't require players to sign in with a PlayStation Network account, and neither will God of War Ragnarok, the Last of Us Part 2 Remastered, and Horizon Zero Dawn Remastered. Okay. Also, the Video Game History foundation opened digital access to more than 30,000 files in its archive of game materials. And there's no playable games, but there's some cool art. And if you're looking for a quick way to get a real search result instead of Google's AI overviews, just make your search sound more angry with some curse words. It'll skip right to the good. Yeah, and you better be here on Monday for more tech news, or so help me, I'll have some choice words for you, like I miss you. Please reconsider your decision.
Episode: Deepseek Flaws, OpenAI O3-Mini, RTX 50 Paper Launch + More!
Host/Author: Linus Media Group
Release Date: February 1, 2025
The episode kicks off with a deep dive into the recent turbulence surrounding the Chinese AI company, Deepseek. While Deepseek hasn't entirely burst the AI bubble, it has significantly impacted Nvidia's stock, causing a "tailspin this week" as investors question the necessity of Nvidia GPUs given Deepseek's "highly efficient" models.
Security Vulnerabilities: TechLinked highlights alarming security oversights by Deepseek. "AI security researchers from the University of Pennsylvania and Cisco found that Deepseek's reasoning model R1 completely failed to protect itself against a barrage of 50 popular chatbot jailbreak techniques," the host explains (00:30). Additionally, Deepseek neglected to secure its internal database of user chat histories and API keys, leaving them "just sitting there without even a password to protect it."
Global Restrictions: These security lapses have led to international scrutiny. Countries including Italy, Taiwan, South Korea, France, and Belgium are actively banning, restricting, or investigating the use of Deepseek models. The host remarks, "multiple countries are banning, restricting or investigating the use of Deep Seq models, including Italy, Taiwan, South Korea, France and even Belgium, who's usually pretty chill" (01:15).
In response to the competitive pressure from Deepseek, OpenAI unveiled its latest reasoning model, the O3-Mini. This new model surpasses its predecessor, the O1, particularly excelling in "science, math, and coding performance."
Enhanced Security Features: The O3-Mini addresses some of the security issues seen with Deepseek by not immediately providing instructions on creating harmful substances when prompted. "It also doesn't immediately tell you how to make cocaine and explosives when you ask nicely, or worse, explosive cocaine," the host notes (02:05).
Access and Pricing: Access to the O3-Mini is tiered. It is available for limited free use through ChatGPT, while "paid users can use it more with $200 ChatGPT Pro subscribers getting unlimited use" (02:30).
Microsoft has been active in leveraging OpenAI's advancements. Currently offering the O1 model via its Think Deeper feature in the Copilot Assistant for free, Microsoft is also embroiled in allegations regarding Deepseek.
Hosting Deepseek R1: Despite questioning Deepseek's use of OpenAI models, Microsoft is "hosting DeepSeek R1 on its cloud services," a move the host humorously comments on: "Keep your friends close and your enemies safely contained in a server farm" (03:10).
Investigations: Microsoft is reportedly "investigating whether deepseek improperly used OpenAI models to train their own" (03:15), indicating a complex relationship between collaboration and competition in the AI sector.
Nvidia's latest GPU launch, the RTX 50 Series, has seen unprecedented demand coupled with significant supply challenges.
Massive Sell-Outs: The RTX 5090 and 5080 models "sold out in minutes at retailers across the US, the UK and probably elsewhere" (04:00). In some European markets, stock appeared virtually nonexistent, attributed to a leaked distributor order link.
Scalper Frenzy: Low inventory and high demand have led to scalpers listing RTX 5090s for exorbitant prices, with examples including listings up to "$9,000 on eBay" and a specific instance where a card was listed for "$5,800 on Facebook Marketplace" by YouTuber Jay Z FittySent (04:40).
Product Value Concerns: Despite the high demand, skepticism remains about the value proposition of the RTX 5080. The host references "a glorious graph made by Paul's hardware" that illustrates the "objectively bad value compared to previous 80 series cards" (05:00).
Amidst the high-demand chaos, Nvidia introduced a new GPU driver feature called Smooth Motion. This spatial frame interpolation tool aims to "bring less powerful frame generation to almost any game," similar to AMD's fluid motion frames.
Exclusive Features: Currently, Smooth Motion is "exclusive to the RTX 50 series," limiting its accessibility to only the latest hardware (05:35). This exclusivity adds another layer of desirability to the RTX 50 Series.
Shifting focus to Intel, the company recently released its financial results, which can be found via the podcast's news sources link.
Product Roadmap Adjustments: Intel announced delays in its product releases, with Panther Lake laptop CPUs expected "later in 2025," and Nova Lake desktop chips postponed until "2026." Furthermore, Intel is "canceling their Falcon Shores AI GPU for now," signaling a strategic pivot away from direct competition with Nvidia's AI accelerators (06:20).
Microsoft's Product Offerings: In a move to stay competitive, Microsoft revealed that its Surface Pro and Surface Laptop for businesses can now be configured with Intel Lunar Lake processors instead of Qualcomm Snapdragon X chips. These configurations come with an additional cost of around "$400" (06:50).
Apple is embroiled in legal maneuvering related to Google's antitrust trial. The company recently filed a motion to delay proceedings after a judge denied their request to intervene.
Financial Dependencies: Apple’s participation hinges on its substantial financial agreement with Google, paying "$20 billion a year to have Google be the default search engine in Safari." The host cynically notes, "without that money, Apple might have to cancel more projects like the Mac Connected AR glasses" (07:30).
AR Glasses Future: Despite rumors about the discontinuation of Mac Connected AR glasses, Mark Zuckerberg from Meta has expressed confidence in the AR market, stating his intention "to sell billions of AR glasses," hinting at a competitive future in augmented reality (08:00).
Wrapping up on a positive note, the podcast highlights several uplifting developments in the tech and gaming sectors.
Account-Free Gaming: Following the launch of Spider-Man 2 on Steam, Sony announced that it "won't require players to sign in with a PlayStation Network account." This policy extends to other titles like God of War Ragnarok, The Last of Us Part 2 Remastered, and Horizon Zero Dawn Remastered, enhancing accessibility for gamers (08:45).
Video Game History Foundation: The foundation has opened digital access to over "30,000 files in its archive of game materials." While these archives do not include playable games, they offer a treasure trove of "cool art" for enthusiasts and historians (09:15).
Search Recommendations: For those seeking genuine search results over AI-generated content, a humorous tip was shared: "make your search sound more angry with some curse words. It'll skip right to the good" (09:40).
Conclusion:
The episode of TechLinked provides a comprehensive overview of significant developments in the tech and gaming industries, from AI advancements and hardware launches to corporate strategies and positive news in gaming accessibility. Whether it's the challenges facing Deepseek and Nvidia, OpenAI's innovations, or Intel and Microsoft's latest moves, listeners are kept informed with insightful commentary and timely information.
Note: Timestamps are approximations based on the provided transcript and are intended for illustrative purposes.