Google Gemini & Microsoft Copilot | 🎙️#38
Why is Google becoming a summary tool? Should you glue your cheese? What future does local AI on device bring? As soon as Kirill has left the building for a week, Pablo and Leo talk AI for 40 minutes: GPT-4o, Gemini and Microsoft Build conference.
- mkdev events this summer;
- How fun is it to talk to AI that can emote in its voice?
- What will happen to Google Search?
- Who is the loser here, the user or Google itself?
- Microsoft’s new hardware for AI on device;
- What to expect from Microsoft Recall?
You can listen to episode 38 of DevOps Accents on Spotify, or right now:
Artificial intelligence continues to evolve at a breakneck pace, and its impact on various technologies, including search engines, is becoming increasingly evident. In a recent discussion, Leo and Pablo delved into several significant topics, including GPT-4o's capabilities, the contentious changes in Google Search with the introduction of Google Gemini, and Microsoft's Recall feature. Here, we explore these themes in detail.
GPT-4o: A New Frontier in Conversational AI
Leo shared his excitement about GPT-4o, particularly its advanced voice features. He tested its ability to simulate real conversations by integrating it into his podcast setup, where GPT-4o could recognize and respond to voices, including his and Pablo's. This seamless interaction highlighted the model's impressive ability to distinguish between speakers and generate contextually relevant responses.
Pablo noted the significant improvement in the naturalness of GPT-4o's voice, emphasizing that it could convey emotions and nuances, making interactions feel less robotic. This advancement mirrors the sophisticated AI depicted in the movie "Her," sparking both excitement and controversy over the ethical implications of such technology.
Google Search and Gemini: A Shift in Internet Interaction
Leo and Pablo disagreed on the implications of Google's latest AI developments, specifically the introduction of Google Gemini. Leo expressed concerns about Gemini's role in transforming Google Search from a tool that connects users to diverse sources of information into a summarizing AI that provides single-perspective answers. He pointed out instances where Gemini's responses were not only incorrect but potentially dangerous, such as suggesting the use of glue to keep cheese on pizza or gasoline for cooking spaghetti.
They had this fancy showcase of Gemini’s capabilities last year, in December, I believe on December 23. Yes, but it turns out the demo was manipulated. They reduced latency for brevity, so to speak, and used still frames instead of live video, like Chat GPT. In essence, the entire thing was fabricated. Yet, Google is now incorporating these AI summaries into everyday search. This is already a significant problem because it scrapes information from actual websites and forces it to the top of the page. This means that Google is choking out a large portion of the internet. For me, this marks a clear transition from curating information to interpreting it for you. Instead of searching for something and browsing it yourself, you are now subjected to hallucinating AI. I don't know who benefits from it except for Google. — Leo Suschev
Pablo acknowledged these issues but argued that the core problem lies in Google's attempt to adapt its business model to the AI era. Google's traditional search model, which relies heavily on advertising revenue from promoted links, is being challenged by AI models like GPT-4o that provide direct answers without leading users to external websites. This shift could undermine Google's primary revenue stream, prompting the company to make hasty, and sometimes flawed, advancements in AI.
The main problem is the way Google is making money because, in this competition, the big loser is Google. Everyone knows that. Amazon is not losing money because they are selling cloud services and packages. NVIDIA is the winner because while everyone else is focused on trivial matters, everyone is buying chips from NVIDIA. They are the happiest company in the world. OK, I love AI even though we don't use AI at all. OpenAI, in the end, is a nonprofit organization, so they don't care about making a profit. That's the reality of this nonprofit company, which is why everyone gave them money.
Microsoft, on the other hand, is jumping from nothing to what it is today because they now have new laptops and computers with pure AI inside. We can talk about that in a few minutes. But OK, everyone is a winner, and the only loser here is Google. The model that Google has been using since 1996 or 1998, for the last 26 years, is the same. There is a tiny box where I type a query, and it gives me a number of links. This model has changed. The way it worked was by introducing ads, generating $175 billion per year. But now this model is starting to collapse. People are still using Google, please don't misunderstand me. But this is the first red flag Google has seen in the last 28 years, a real red flag, signaling that they could be in trouble. — Pablo Inigo Sanchez
Microsoft Recall: Enhancing Convenience with Privacy Concerns
Another topic of interest was Microsoft's new Recall feature, part of their Copilot AI integration in Surface computers. This feature continuously saves and indexes all activity on the device, allowing users to search for past actions and information effortlessly. Leo praised the convenience of Recall, highlighting its potential to revolutionize how users interact with their devices by enabling instant access to past activities and data. However, Pablo raised valid concerns about the privacy implications of such a feature. While Microsoft claims that Recall does not index DRM-protected content or sensitive data like credit card information, the idea of having all user activity recorded and searchable poses significant security risks. The balance between convenience and privacy remains a critical issue as AI technologies continue to develop.
At the conference, Microsoft unveiled new hardware, specifically a Surface computer with ARM chips similar to those in iPads. However, this device resembles modern Windows laptops more than iPads. According to Microsoft's measurements, these devices are claimed to be significantly faster than the latest MacBook Air, which is questionable because these claims are often based on comparisons with lower-end Apple models.
More importantly, they introduced a product called Copilot, an AI-based feature integrated into these computers on an autonomous basis. This Copilot requires a specific chip to function quickly, and I’m not sure if it requires an internet connection. The most controversial idea is that the model is secured inside the chip, making it local. This is significant because, as we've discussed in previous podcasts, the next step for artificial intelligence will be operating locally on devices. This is what Microsoft is aiming to do.
The most controversial aspect of Microsoft's Copilot is a feature called Recall. This feature continuously runs in the background, saving everything happening on the device, including applications used, text written, and images viewed. It indexes this information locally, allowing users to search for any past activity. From a security perspective, this sounds frightening, but in terms of convenience, I think it's incredible. — Leo Suschev
Navigating the AI Landscape
The discussions between Leo and Pablo underscore the complexities and challenges of integrating advanced AI into everyday technologies. While innovations like GPT-4o and Microsoft Recall offer exciting possibilities, they also raise important questions about privacy, security, and the future of information dissemination. As AI continues to evolve, it is crucial for developers, users, and policymakers to address these concerns to ensure that the benefits of AI are realized without compromising ethical standards and user trust.
Show Notes:
- Pablo’s video on Google’s future
- Microsoft Build conference
- ICO investigates Microsoft over Recall feature
Podcast editing: Mila Jones / milajonesproduction@gmail.com