Google Is Back, OpenAIs 12 Days of Demos, and the Deal Book Summit – Live and Learn #57

✨ Quote ✨

To see the world, things dangerous to come to, to see behind walls, draw closer, to find each other, and to feel. That is the purpose of life.

– James Thurber (from The Secret Life of Walter Mitty)

✍️ (Traveling) Posts ✍️

I spent some time writing up the adventures from Martinique and Portugal, and you can check everything out, with pretty pictures in the travel section of this website. I've also been thinking about splitting out traveling-related content from this newsletter into its own separate thing. But I am not sure whether or not or how to do it yet.

🖇️ Links 🖇️

Google has released so many crazy things in these last two weeks... it's insane. Gemini 2, Project Mariner, Deep Research, Project Astra, Genie 2, and Deepmind Weather Forecasting?! Just wtf. I'll provide more details on what each of these does below, but in general, it feels like things are accelerating drastically, both at Google and OpenAI. It feels like AGI is in the air and as if the compounding effects of technological advances are starting to pay off and become extremely visible. And it makes me really wonder what 2025 is gonna be like.

A lot of things have happened that I want to share here, and it would not give the full picture if I omitted too much these two weeks. Part of the progress of AI is just how much stuff is happening. I'm already omitting 150+ other links that I've found and skimmed or read, and yet here we are, with a crazy full two weeks of research results and announcements. So, without further ado, here's what happened in the world of AI, just grab a drink and settle in 😊

Gemini 2.0 by Google. The newest version of Gemini is Google's LLM for powering agentic workflow. It has powerful multi-modal capabilities and outperforms other SoTA models on many benchmarks. In the words of Sundar Pichai: "If Gemini 1.0 was about organizing and understanding information, Gemini 2.0 is about making it much more useful." Words that ring true to Google's mission. The announcement video is worth watching too, and you can try out the model at Googles AI studio. Their gaming demo with Gemini 2.0 was wild. It's like having a friend watching over your shoulder, helping you play the game, and explaining the game mechanics and meta while you play it in real time. The main Gemini page also has more information on the model. Gemini has spatial understanding, can read and help you with maps, and has real-time video capabilities. It can also output text and audio, plan, reason, and even help you execute actions. All of this is across different languages, of course. On top of all that, they tried to make it easy for developers to extend, use, and build with the new Gemini 2.0.

Because of this, Gemini 2.0 powers a bunch of new tools, ideas, and use cases.

The Developer Updates alone are nuts. Jules, an agent that can help with code, a Data Science Agent running in Google Colab, and API integrations for tool use, real-time multi-modal interactions, and spatial understanding exist. People are building wild stuff integrating with Gemini 2.0 already, like tldraw, viggle, or rooms.

Then there is Deep Research, which uses Gemini 2.0 under the hood to be able to scan hundreds of documents across the web to answer your query. It is like a research assistant doing an in-depth literature review across the entire web and then synthesizing across hundreds of sources to provide a comprehensive analysis and answer to your question. It's crazy, and you can try it out on the Gemini Web App.

They also teased Google Project Astra–an AI real-time video-based assistant that can see and hear everything through your phone, and can help you navigate the world, give real-time information about everything you encounter and even works across different languages. This seems like magic, even in today's crazy times. You should totally watch their Project Astra demo video, but for a more balanced and realistic review, without the marketing fluff watch this test by Matt Wolfe.

And finally, Google also released a sneak peek into Project Mariner–a Chrome extension that can navigate the web for you. It's like a personal assistant that can do research, buy things, plan your day, post to social media, and do a large variety of tasks that you'd normally do online. At the moment, it is still a bit slow and sometimes inaccurate, but that is bound to improve. They have a video demo of what it's like to use it, but right now, it is only available to a select few testers. You can join the waitlist by filling out this Google form though.

Google was not done cooking these past two weeks, though. Besides the releases for Gemini, they also released Genie 2, Willow Quantum Chips, and their Deepmind Weather Forecasting model. To me, it's crazy that one company announced all these updates in two weeks, but this is only a testament to Google's insane scale and competitive excellence.

Genie 2 by Google Deepmind. Google is pushing hard to make real-time video game AI models a reality. And their second version of Genie is a huge step in this direction. It improves tremendously on the prior version of Genie by adding 3D capabilities and up to one-minute playable demos. The rate of improvement is insane, and I wonder how long it takes before we see the first real games entirely generated by AI.

GenCast Weather Predictions by Google Deepmind. This improved model can predict the weather accurately up to 15 days ahead. It beats the current systems with higher accuracy. Still, weather is impossible to predict perfectly, so the model doesn't provide just one forecast but instead gives an ensemble with likely scenarios that might happen, weighted by their probabilities. It was trained on a bunch of multi-decade weather data, and "the model learned global weather patterns, at 0.25° resolution, directly from this". You can read the paper here.

Willow Quantum Chips by Google. Google also released an announcement for their new quantum chip named Willow. It improves drastically upon quantum error correction and increases the possible number of stable qubits working together. The big deal is that their quantum error correction gets better as they scale up the number of Qbits, not worse. This is something that has never been done before. Before, the bigger you made the system, the noisier and error-prone it got until it quickly became unstable and unusable. With Willow, this seems to have changed now, which is a crazy breakthrough. In the words of the authors, Willow "is the most convincing prototype for a scalable logical qubit built to date. It’s a strong sign that useful, very large quantum computers can indeed be built." They showed that Willow could do a computation that normally would take 10^25 years on a current classical supercomputer, and the people at Google think that this gives credence to David Deutsch's argument for the existence of the multiverse that is carrying out the computation. This is wild stuff.

Human Torso Android Robot by Clone Robotics. This company is speed-running to make West World a reality. I feel that their Android human-robot looks like the freaking intro from West World on purpose. And now you can even buy the first few prototypes they produced to help you do all kinds of household tasks. It's a bit scary to see them building literal android robots with fully biomimetic muscles, skeletons, and everything. I mean, these things look human-like and move human-like to a scary degree...

12 days of demos by OpenAI. The 12 days are not fully over yet, but the announcements have been interesting so far. The biggest was their 200$ per month premium plan that gives you full unlimited access to o1 and opening up Sora to the public. Marques Brownlee did a great review on Sora. I love his style and balanced assessments of Sora's capabilities. To me, it feels like there are already stronger video models out there, though, like the Chinese Hunyuan Video created by Tencent. One thing is for sure: AI Video is here to stay, and it's the worst it has ever been.

Deal Book Summit by New York Times. This year's deal book summit again had a bunch of really good interviews. I enjoyed those with Sam Altman, Jeff Bezos, and Sundar Pichai the most. The questions that Ross Sorkin asks are often pointed and hard to answer, but therein lies the beauty of his interview style. He's good at getting to the bottom of often controversial topics and not afraid to ask the difficult questions.

Here are a few more quick links: Fei Fei Li's World Labs has released its first update. Amazon has finally revealed its own foundation model called Nova LLM. And Devin, a complete AI-based junior-level software engineer is out and generally available. It costs 500$ a month, and the opinions on X are sort of mixed, mainly because of the price tag. But it seems genuinely useful, though.

🌌 Travel 🌌

The last two weeks of traveling in Colombia have been super exciting. I successfully found and bought a touring bicycle and fitted it out with all the saddlebags, spare parts, and things I'll need for this long trip. Now, I have started cycling around Colombia but quickly got stuck because I enjoyed my time too much at a little Finca.

It's a cute little paradise here with the chickens, dogs, and cats living in the middle of the jungle, a bit South of Fusagasugá. It's so tranquil here that it nourishes my soul, and the planned stay of one night turned into a week. And I have to make sure not to get stuck here for too long because it's that beautiful.

Right now, life is simple: working a little bit from my computer, drinking homegrown coffee while listening to the birds in the trees and Jazz music, and watching the Colibri in the hibiscus flowers. Eating arepas, fresh avocado, oranges, and soursop (a local fruit) from the trees, eggs from the chickens of the farm... It just feels soooo wholesome that I am really not sure how long I want to stay here before finally continuing. I love my freedom and that I do not have to go anywhere.

Here are some glimpses and impressions from the last weeks in Bogotá, as well as the tranquil farm and the hunt for the bicycle.

🎶 Song 🎶

Meet Me There by Nick Mulvey

I've been listening to a lot of music by Nick Mulvey lately and loving the calm, slightly sentimental, and sad mixed with hopeful mood. It's relaxing to me, and I play his music on repeat. Check him out!

Youtube Music | Spotify

---

That's all for this time. I hope you found this newsletter useful, beautiful, or even both!

Have ideas for improving it? As always please let me know.

Cheers,

– Rico