OpenAI o1, Carpentopod and Diatom Art – Live and Learn #51

✨ Quote ✨

 I believe that we're now entering a phase in startup history when building a $1m ARR business with a 5 people team is going to be not just possible, but easy. An environment of low defensibility and high consensus produces a fragmentation of entrepreneurship, as opposed to winner-take-all outcomes, which can be either amazing (for lifestyle founders) or bad news (for big funds).

– Gian Segato - (source)

I think this quote reflects the times that we are heading into, with tools becoming so powerful that the efficiency of single people to create and add value to the world by using these tools is increasing drastically. But I think there is also a point that this quote misses: At some point in time, the tools might become so capable that they don't need the human in the loop anymore. There will be 0 people teams producing $1m+ ARR businesses in the future. And that's a pretty wild idea to wrap my head around.

✍️ Post ✍️

Diatom Art Arrangements.

I collected a bunch of microscopic diatom art arrangements from the internet and put it all together in this blog post. Now I have something of intricate beauty to show people and blow their minds. Check it out!

🖇️ Links 🖇️

o1 launch by OpenAI. This is the biggest news of the last two weeks by far. Heck, it might be the biggest announcement of the entire year. OpenAIs o1 is an LLM that can spend more inference compute time to think and reason about a problem step by step before giving an answer. I think what we see here is a fundamental shift in how useful LLMs can be. It seems like OpenAI has found a way of pumping the little speck of proto-intelligence inherent in an LLM like GPT-4 over and over again to produce something that feels different. o1 thinks and reasons much closer to how we do. I have tried out the o1-preview for a side project I am coding right now and so far I have been super impressed by how well it handles this sort of task. It is subjectively much better at it than GPT-4 or Claude Sonnet. But more importantly, this is the worst that it will ever be and I think from here on out this sort of chain of thought reasoning will become much more prevalent and better. For any sort of task that needs planning or thinking ahead this model makes a huge difference. It truly feels like we are heading towards a world where AGI is imminent. OpenAI released a bunch of content around this launch too. Their o1-preview and o1-mini are available in ChatGPT. And they have a whole playlist of videos on o1's capabilities. All of them are worth watching. They also released this post and paper with more details on how they achieved all of this and the benchmarks of this new model. All in all this tweet by Noam Brown, one of the researchers on the Strawberry team behind o1, sums up the significance quite well: "@OpenAI's o1 thinks for seconds, but we aim for future versions to think for hours, days, even weeks. Inference costs will be higher, but what cost would you pay for a new cancer drug? For breakthrough batteries? For a proof of the Riemann Hypothesis? AI can be more than chatbot". For more analysis on the topic, you can also check out these two blog posts by oneusefulthing and lifearchitect. I recommend reading all of this because this is important stuff happening right now that might impact the future of humanity.

Carpentopod - a Walking Table by Giliam de Carpentier. I love it when nerds do cool shit just for the fun of it and this is a prime example of that. Building a custom leg joint that can move a table around without the surface of the table moving is a hard engineering problem... And then building this system in real life is even more impressive. And it just looks so trippy, like some sort of magical artifact. It's projects like this that keep inspiring me to do more, to learn more, and to be curious about the world and how to build epic things within it.

World Labs Startup by Fei Fei Li. Fei Fei Li is one of the legends of the AI world and taught many of the best researchers in the world. Seeing her announce a new startup feels good and their goal of trying to create a new set of World Models that have spatial intelligence around the 3D world is inspiring. If they succeed, this could be quite big for translating the benefits of AI into the real world. In a way, this is fundamentally about making AIs better at physics, robotics, and engineering => in the end, tasks that have real-world implications.

Can LLMs generate Novel Research Ideas?. This is a pretty long paper but the gist and main take-away of it is this: LLMs can generate novel research ideas better than humans can. This, combined with more advanced reasoning like o1's capabilities could lead to models that are exploring and then answering their own research questions in an open-ended scientific loop: posing and then answering questions about the universe in a way that is only bound by the amount of compute resources that you can throw at it. We're in for a wild ride.

Language Agents Achieve Superhuman Synthesis of Scientific Knowledge by FutureHouse. On a similar note, AIs can also do literature reviews and synthesize knowledge across different research papers. To me, this combined with the previous idea is a very big deal. It means we could have an AI that is summarizing the research findings of other AIs making the core insights of the scientific loop accessible to humans. In the words of the authors of this paper: "PaperQA2 writes cited, Wikipedia-style summaries of scientific topics that are significantly more accurate than existing, human-written Wikipedia articles." I repeat... We're in for a wild ride.

🌌 Traveling 🌌

I am still at the Garden enjoying life. The forest fires have been raging around Portugal but we were safe and sound in our little lush forest here. Luckily.

I have been working on a few side projects here and there, like the Diatom post, but mostly was just enjoying the company of the people around me. One of the coolest things that I got to experience was a demo of the Apple Vision Pro. A friend here has one and showed it off to people over the last few weeks. It's an insanely cool device and my brain literally just forgot that it was wearing this thing and accepted the see-through video screen as reality. This to me was pretty mindboggling, and the experiences that can be built on top of it will be even more so. VR content is going to be a new form of media and art I think.

🎶 Song 🎶

Sometimes I Like to Lie by annenmaykantereit

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