Coffee of the Week
The AI Kingdom is on fire! A summary of the main news of the week
Hello everyone! I've put together a hot roundup of the most interesting news about Artificial Intelligence this week. There's so much going on that it's hard to keep up! Let's dive into this fascinating universe and find out what goes on behind the scenes at the biggest tech companies.
Meta responds to Google with an open source version of NotebookLM
Meta has just released NotebookLlama, an open source version that competes directly with Google's NotebookLM. The project is a series of notebooks/tutorials that allow you to create an interesting workflow for converting PDFs into podcasts. The system uses different versions of the Llama model (1B, 8B and 70B) for different steps of the process, from PDF pre-processing to creating dramatic podcast transcripts.
It's interesting to see how Meta is approaching the competition with Google. Instead of simply creating a proprietary product, they've opted for an open source approach that allows developers to not only use, but also modify and improve the system. What’s even more interesting is how they structured the project to work with different model sizes, making it accessible even to those without access to powerful GPUs. A smart move that shows how competition in the AI market can benefit the entire development community.
Meta plans to revolutionize search with AI
Meta is developing an AI-powered search engine that promises to compete directly with Google. The project is still in development, but the proposal is to create a more conversational and contextual search experience, leveraging the power of large language models.
Google has dominated search for so long that it was about time someone shook up the market. Meta’s approach, which now joins Perplexity and OpenAI in focusing on conversational AI, could be just the thing needed to create a viable alternative to Google. But will it be enough to change such ingrained user habits? Here at Yes, we've already adopted Perplexity extensively!
Perplexity AI arrives on macOS
Perplexity AI, an AI-powered search engine that has been gaining a lot of attention, has just launched its native application for macOS. The tool promises smarter searches and more accurate answers, with an interface perfectly integrated into Apple's operating system.
As a Mac user (just kidding! 😅), I can say that the arrival of Perplexity is excellent news. The integration with the operating system is impeccable, and the user experience is much more fluid than using the web version (at least that's what I imagine).
ChatGPT now has advanced search capabilities
OpenAI is not messing around! ChatGPT now has advanced search capabilities that allow users to find specific information in their past conversations. In addition, the tool can now search for up-to-date information on the web, further reducing its dependence on Google.
This update shows how OpenAI is focused on making ChatGPT an increasingly complete tool. The ability to search through past conversations is particularly useful for those who use ChatGPT professionally, like me who share the account with my development colleagues. It's like having an assistant that not only helps you, but also reminds you of all your past conversations.
Google Maps Embraces AI with Gemini
Google Maps is getting even smarter with the integration of Gemini AI. The app can now answer complex questions about locations and suggest directions in a more contextual and natural way.
This update shows how AI is seeping into every aspect of Google’s services. The integration feels natural and genuinely useful — it’s not just another AI feature thrown in for marketing. It remains to be seen how honest the AI will be in answering, or will it also offer sponsored answers?
OpenAI’s advanced voice is now free (but with mimicry)
OpenAI is making its advanced voice functionality free for 10 minutes a month.
Despite the limited time, it’s a smart move by OpenAI. Ten minutes may not seem like much, but it’s enough to understand the potential of the technology and decide whether it’s worth investing in the paid version. It’s like that free sample that makes you want to buy the full product!
OpenAI surprises with revenue numbers
A surprising revelation: 75% of OpenAI’s revenue comes from paying customers! It’s an impressive number that shows how the company has managed to create a product that people truly value and are willing to pay for.
These numbers are even more impressive when you consider that most tech companies rely primarily on advertising or corporate revenue. OpenAI has managed to do something rare: convince individual users to pay for an AI service. It’s a testament to the real value that ChatGPT offers its users.
The Whisper Dilemma in Hospitals
A recent study found that OpenAI’s transcription system, Whisper, sometimes “hallucinates” in medical transcriptions, raising concerns about its use in hospital settings.
This case perfectly illustrates the dilemma we face with AI: balancing efficiency with accuracy and security. On the one hand, Whisper can greatly speed up medical documentation work; on the other, any mistake can have serious consequences. It’s a clear example of how cautious we need to be when deploying AI in critical sectors.
OpenAI enters the chip world
OpenAI is developing its own chips in partnership with Broadcom and TSMC. It’s an interesting move that shows how the company is seeking greater control over its technology infrastructure.
This decision by OpenAI is curious for several reasons. First, it shows how the company is becoming more verticalized. Second, it indicates that they are looking to optimize hardware specifically for their AI models. It’s as if they are saying, “If we want to do this well, we need to control the entire process.” The problem is that this will end as a monopoly.
Claude on the desktop: It's not all sunshine and roses
Anthropic has released a desktop app for Claude, but not all the features of the web version are present. It's an interesting first step, but it leaves some users wanting more.
Look, another one landing on the computer! Is this a trend? As a regular Claude user, I confess that I was a little disappointed with the limitations of the desktop app. On the other hand, it's understandable that Anthropic would want to launch a leaner product first and gradually add features.
Meta and the 100,000 NVIDIA GPUs
Meta is using more than 100,000 NVIDIA H100 GPUs to train Llama 4. It's an astronomical number that shows the massive investment the company is making in AI.
To get an idea of the magnitude of this investment: each H100 GPU costs thousands of dollars. We're talking about a billion-dollar investment in hardware alone. Will this translate into a model that is truly superior to its competitors? We win, but nature loses with the excessive consumption of energy...
An official definition for Open Source AI
We finally have an official definition for what constitutes a truly open source AI! This standardization was much needed to avoid misuse of the term and ensure greater transparency in AI development.
This definition is an important milestone for the AI community. We now have clear criteria to evaluate whether a project is truly open source or if it is just using the term as marketing. It's like having a quality seal for open source AI projects.
Google's Project Jarvis
Google is developing Project Jarvis, an AI system that promises to automate complex tasks on a computer. It's like having a virtual assistant that actually understands how to use a computer.
This project reminds me a lot of Iron Man's assistant, and it's no coincidence that they chose the name Jarvis. The idea of having an AI that can actually interact with user interfaces is fascinating. I just hope that, like in the comics, Jarvis doesn’t become a conscious and become a villain…
GitHub Copilot is now multi-model
GitHub Copilot is expanding its horizons! The tool now supports multiple AI models, including Anthropic’s Claude and Google’s Gemini.
This update is particularly interesting because it shows a trend of openness in the AI market. Instead of being locked into a single model, Copilot is transforming itself into a platform that can make the most of every model available. It’s like having multiple experts working together on your code.
Apple enters the AI game
Apple has finally revealed its cards in the AI game, the company is launching artificial intelligence features integrated into iOS, iPadOS and macOS. As you would expect from Apple, the focus is on privacy and seamless integration with the company's ecosystem.
Apple's entry into the AI market was expected, but the approach is typically Apple: focused on practical use cases and deep integration with the operating system. It's funny to see how they're avoiding the term "AI" in favor of "Intelligence", probably to distance themselves from the current market hype.
The AI race is getting fiercer, with every major tech company looking to differentiate itself. What strikes me most is how, in just a few weeks, we’ve seen so many significant advances. From OpenAI building its own chips to Apple finally getting in on the action, it’s clear that AI isn’t just a passing fad—it’s the future of technology unfolding before our very eyes.
What’s fascinating is how different companies are approaching similar challenges in different ways. While some are focusing on custom hardware, others are focusing on strategic partnerships. Some are prioritizing privacy, others are focusing on openness. In the end, the winners are us, the users, who have more and more options and powerful tools at our disposal. But I still think that if we don't think about energy compensation and environmental compensation, this future may not be so wonderful.
