Google IO, GPT4-o and AlphaFold 3 – Live and Learn #42

✨ Quote ✨

When in doubt, follow your curiosity. It never lies, and it knows more than you do about what's worth paying attention to.

– Paul Graham - (source)

🖇️ Links 🖇️

GPT-o announcement by OpenAI. OpenAI came out with a big announcement that is not yet GPT-5 but is still a big step forward. GPT4-o is a multimodal LLM with excellent voice in- and output capabilities. Using it feels like talking to a real person–especially the AI tutor and coding assistant demos seem extremely promising. We are very close to the idea of the Primer from Neal Stephenson's book "The Diamond Age". Best of all, OpenAI plans to make GPT4-o available to the public for free.

Google I/O Keynote by Google. Google I/O also happened in the last two weeks. It's been all about AI once again and the biggest announcements were the 1 million token context window of their Gemini 1.5 models, Imagen3 their newest image generation AI, Astra their personal assistant project, and Veo their new video generation model. It seems like Google is fighting hard not to lose to OpenAI, and it seems like Google can at least match what OpenAI is building. But let's see what happens once GPT-5 drops later this year. For now at least Astra looks more powerful and nicer than GPT4-o from the demos, but this impression might change once people get access to both models.

Alphafold 3 by Google. To me, this is one of the biggest things that happened in the last two weeks. AlphaFold 3 can not only model the structures of amino acid chains folding up into complicated proteins–it can also predict the protein's interactions with other biomolecules such as RNA and DNA. This is huge for drug design and exploration. Again, it's such a big step up from AlphaFold 2. And best of all, it's accessible for researchers online. This makes it a true game-changer.

Interview, with Sundar Pichai from The Circle. I really enjoyed this interview with Sundar Pichai, which shows that Google is trying to make long-range bets work while staying true to its mission of organizing the world's information. Nobody is sure how they are going to monetize the AI-generated search content they are pushing for, but they push for it nonetheless. The interviewer asked lots of challenging questions, and Sundar answered them with a kind of corporate grace, which was delightful to see.

Fully Body Haptics Demo by University of Chicago. This paper introduces a way to enable full-body haptics for VR and AR content. I thought the idea they had was cool: wear a hat with a TMS device that can move. TMS stands for trans cranial magnetic stimulation, and it's essentially a way of stimulating the brain from outside the skull using strong electromagnetic pulses. There are still shortcomings in their approach, though: the biggest one is that you can't stimulate multiple areas at the same time. The paper page can be found here.

🌌 Travel 🌌

I've been traveling around Dominica for the last two weeks, and my time here is flying by so quickly that it's scaring me. But the places are tremendously serene, and the nature is overwhelming... and I clicked waaaay too many pictures. There are so many waterfalls, geothermal springs and beautiful beaches. Dominica is truly out of this world. And it's Mango season right now and the people are immensely friendly, so I've been living a really good life.

🎶 Song 🎶

Iron by Woodkid

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