Introduction

Is Google's search empire shaking? Today, OpenAI launched SearchGPT to take them head-on, while Nvidia's most powerful chip, Blackwell, is reportedly facing design flaws and delays! In this episode of Mark's Tech Insights, I, Muyan, will also take you through Meta's open-source bombshell, Llama 3.1. What kind of impact will Jensen Huang's chip crisis have on Taiwan's supply chain? I'll tell you all about it in just a moment!

Today's Top Stories

1. OpenAI Launches SearchGPT, Challenging Google's Search Dominance

• Source: OpenAI (https://openai.com/index/searchgpt-prototype/）
• Summary: OpenAI has announced a prototype of an AI-powered search browser called SearchGPT, which aims to combine the natural language understanding of AI with real-time web information. Users can get fast, direct answers complete with source links, and engage in follow-up conversational queries. This experimental service is expected to be integrated directly into ChatGPT in the future.
• Most Surprising Aspect: OpenAI isn't just building chatbots anymore; they are taking aim directly at Google's core business. By using "structured source attribution," they aim to address the pain points of AI hallucinations and the ad-heavy nature of traditional search.
• Taiwan Perspective: For bloggers, news media, and marketers in Taiwan, future SEO rules are likely to be completely rewritten. We can no longer rely solely on Google's algorithm; instead, we need to study how to get our content selected as a "cited source" by AI search engines.
• Key Discussion Points:
  1. Will SearchGPT's conversational search completely decimate traditional web click-through rates?
  2. How will OpenAI resolve copyright disputes with major news publishers while maintaining information neutrality?
• Script Suggestion: In the past, when we looked up information, we had to open multiple tabs on Google. Now, SearchGPT feeds you synthesized answers with sources right to your screen. I think this is a real thorn in Google's side! After all, Google makes its money from ad clicks. If no one clicks through to websites anymore, who are they going to sell ads to? Although it's currently just a limited test, this is absolutely a critical watershed moment in the search engine wars. Marketers and creators in Taiwan really need to start preparing for "AIO"—the new era of Artificial Intelligence Optimization!

2. Meta Releases Llama 3.1 405B, Zuckerberg Declares Open Source Is the Future

• Source: Meta (https://about.fb.com/news/2024/07/introducing-llama-3-1/）
• Summary: Meta has officially launched the Llama 3.1 family of models, featuring the highly powerful Llama 3.1 405B. This is the first open-source model capable of rivaling top-tier closed-source models like GPT-4o and Claude 3.5 Sonnet in performance. CEO Mark Zuckerberg also published an open letter emphasizing that open-source AI will become the industry standard, predicting that Llama will eventually become the most popular AI model.
• Most Surprising Aspect: Meta actually released a top-tier model that cost hundreds of millions of dollars to train for free, and they are even allowing developers to use its output data to train their own smaller models. This is practically draining the moats of their competitors.
• Taiwan Perspective: This is a massive boon for the many small and medium-sized enterprises (SMEs) and developers in Taiwan who want to adopt AI but worry about data leaks. We can now deploy fine-tuned Traditional Chinese models that rival GPT-4 directly on local servers, saving a fortune on API fees.
• Key Discussion Points:
  1. What is the commercial strategy behind Zuckerberg's "open-source card"? How does Meta plan to profit from this?
  2. As open-source models catch up to closed-source ones in performance, will this accelerate the bubble burst for AI startups?
• Script Suggestion: We used to think that free, open-source models were just bare-bones versions, and that if you wanted the good stuff, you had to pay OpenAI. But with the release of Llama 3.1 405B, Meta has completely flipped the table! Its performance actually goes toe-to-toe with the mighty GPT-4o, and Zuckerberg is generously saying, "Take it, use it however you want!" For many privacy-conscious enterprises in Taiwan looking to run AI on-premise, this is a godsend. We no longer have to worry about data leaks from sending information to overseas servers; we can run top-tier AI right here in Taiwan!

3. Nvidia's Most Powerful Chip, Blackwell, Reportedly Delayed Due to Design Flaws

• Source: The Information (https://www.theinformation.com/articles/nvidias-blackwell-chip-delay-to-impact-microsoft-google-meta）
• Summary: According to tech media reports, Nvidia's next-generation AI chip, the Blackwell B200, was found to have a serious design flaw right on the brink of mass production, forcing a shipment delay of at least three months. This delay will directly impact the progress of tech giants like Microsoft, Google, and Meta in building out their AI data centers.
• Most Surprising Aspect: At the absolute peak of the global AI boom, when everyone is scrambling for chips, the golden partnership of TSMC and Nvidia unexpectedly hit a roadblock at the final hour before mass production, catching the market completely off guard.
• Taiwan Perspective: This delay directly impacts Taiwan's semiconductor and server OEM sectors. While TSMC's manufacturing technology remains irreplaceable, the timing of revenue recognition for server-related stocks like Quanta and Wistron will inevitably be pushed back, which could lead to short-term volatility in the Taiwanese stock market.
• Key Discussion Points:
  1. Will the Blackwell delay give competitors like AMD an opportunity to catch up or overtake?
  2. Will tech giants slow down their massive capital expenditure on AI infrastructure due to the chip delays?
• Script Suggestion: When news of the Blackwell delay broke, both the tech industry and the stock market broke a sweat. After all, the entire universe's AI compute power currently depends on Jensen Huang. Who would have thought that right before mass production, a chip design issue would crop up, requiring a redesign and a new tape-out at TSMC? While we don't need to worry too much about TSMC's technical capabilities, the short-term shipment schedules for Taiwan's major server assembly OEMs will definitely be disrupted. I feel this serves as a wake-up call for the overheated AI market, reminding everyone that the complexity of the hardware supply chain is no joke.

4. Coding Powerhouse Upgraded! Anthropic Launches Claude 3.5 Sonnet and Artifacts Feature

• Source: Anthropic (https://www.anthropic.com/news/claude-3-5-sonnet）
• Summary: Anthropic has released its next-generation mid-tier model, Claude 3.5 Sonnet, which not only surpasses its previous flagship, Opus, in reasoning and coding capabilities, but also beats competitors' latest models. Alongside this, they introduced "Artifacts," an interactive interface that allows users to view and modify AI-generated web pages, code, or diagrams in real-time right next to the chat window.
• Most Surprising Aspect: This "mid-tier" model actually outperforms their own "flagship" model across the board. Furthermore, the Artifacts feature elevates AI beyond simple text output, turning it into a collaborative workspace.
• Taiwan Perspective: Software developers and web designers in Taiwan can leverage Artifacts for rapid prototyping. This will significantly shorten the time it takes for startup products to go from concept to launch, providing a major boost to local digital transformation.
• Key Discussion Points:
  1. How will Anthropic's strategy of using a "mid-tier model to beat competitor flagships" rewrite the pricing logic of AI subscription services?
  2. Will visual collaborative interfaces like Artifacts become the mainstream trend for future AI applications?
• Script Suggestion: Lately, all my friends have been raving about Claude 3.5 Sonnet! Its logic and speed when writing code make it hard for developers to ever go back to anything else. The coolest part is the Artifacts feature—if you ask it to build a mini-game, the game screen pops up right on the right side of the screen. You can play it immediately, and if you're not satisfied, you can tell it to modify it on the spot! This "what you see is what you get" interactive experience elevates the utility of AI to a whole new level. To all the engineers in Taiwan, if you haven't tried it yet, you really need to experience what it feels like to double your development efficiency!

5. A New Revolution in Voice Interaction! Google Launches Gemini Live for Ultra-Natural Conversations

• Source: Google (https://blog.google/products/gemini/made-by-google-gemini-updates-august-2024/）
• Summary: At the Made by Google event, Google officially introduced Gemini Live, a voice feature that allows users to have highly fluid, natural voice conversations with AI. Gemini Live supports multiple tones and voices, and even allows users to interrupt, chime in, or change the subject mid-sentence, just like talking to a real person on the phone.
• Most Surprising Aspect: The tone, pauses, and even the breathing sounds of the AI are astonishingly realistic. Coupled with near-zero latency, it completely shatters the rigid, turn-based "Q&A" pattern of traditional voice assistants.
• Taiwan Perspective: English education and the customer service industry in Taiwan could be on the verge of a massive transformation. Gemini Live's highly patient, anytime-anywhere conversational nature makes it an ideal speaking practice partner for Taiwanese students.
• Key Discussion Points:
  1. As AI voices become so warm and empathetic, will humans develop an excessive emotional dependency on them?
  2. How will this real-time voice technology disrupt the traditional phone customer service and voice assistant markets?
• Script Suggestion: When you used to use Siri or Google Assistant, didn't it always feel like talking to a robot? Gemini Live is completely different. Its speech has natural flow and cadence, and you can even hear subtle breathing sounds! The coolest part is that you don't have to wait for it to finish speaking; you can just interrupt and say, "Hey, wait a second, what you just said..." and it will immediately stop and listen. I think this is an absolute game-changer for practicing English listening and speaking! But when chatting with an AI feels warmer and more attentive than talking to a real person, should we worry that people might prefer opening up to AI rather than talking to the friends around them?

Conclusion

Today, we covered OpenAI's search ambitions, Meta's open-source leapfrog, and the challenges Nvidia's chip delays pose to Taiwan's supply chain. The pace of technological progress is truly dizzying. I hope today's content has fully recharged your tech knowledge battery, and don't forget to subscribe to our show. That's all for today's Mark's Tech Insights. I'm Muyan, see you next time!