Planning for AGI, Code that Fixes Itself and SAM – Live and Learn #15
Welcome to this edition of Live and Learn. This time with a system named Wolverine that enables code to fix itself, an article describing the implications of a new paradigm called Software 3.0, and OpenAIs take on how they work towards creating AGI, while making sure that it's beneficial for as many people as possible.
✨ Quote ✨
If somebody builds a too-powerful AI, under present conditions, I expect that every single member of the human species and all biological life on Earth dies shortly thereafter.
– Eliezer Yudkowsky - (source)
🖇️ Links 🖇️
Segment Anything Model (SAM) by Meta AI. Segmentation is the task of finding the objects in an image and then making this identification usable for the computer. Which part of the image shows what, where are the edges of things, etc. This crazy model released by Meta is very very good at this task of segmentation. Like the title of the paper suggests it can segment anything. This is huge, especially for augmented reality which is why Meta puts a lot of effort into researching these things.
Wolverine by BioBootLoader. A Python interpreter which enables "self-healing" code. WTF. You write code, it has bugs, and GPT-4 finds and fixes those bugs until you have bug-free code. It is, essentially, an automatic debugger. Even crazier, hooking this up to a test suite will soon enable an extreme version of test-driven development. Define what you want as a bunch of test cases, and let GPT-4 figure out the rest, recursively, until the program passes all your test cases, with only minimal human oversight.
Software 3.0 by Divyansh Garg. This is the logical next step after Software 2.0, a famous idea by Andrej Karpathy. Instead of building your own, you use already existing models to build software systems using only plain English. Put in the requirements, and get out the finished product. And with tools like the above and AutoGPT, this vision of Software 3.0 is already here. Soon everybody can build arbitrarily complex software, without knowing about any of the lower-level details. Just as people don't need to know Assembly to write complex websites, Android apps, or 3D games, people in the future won't need to know the syntax of programming languages or APIs and similar paradigms anymore. We will be able to transcend those concepts and work in yet another layer of abstraction.
Planning for AGI and Beyond by OpenAI. It's really interesting to see that OpenAI puts a lot of thought into what they are doing, why they are doing it, and how they think the world is going to change because of their actions. Their mission statement is wonderful: "Our mission is to ensure that artificial general intelligence—AI systems that are generally smarter than humans—benefits all of humanity." This post dives into what that means, and how they plan to get there. On the same topic, there is a way longer Lex Fridman interview with Sam Altman, one of the co-founders of OpenAI. It is an interesting back-and-forth conversation, teasing out the philosophy of this company and its founders.
🌌 Midjourney 🌌
🎶 Song 🎶
Greenpath by Christopher Larkin
I hope you found this newsletter useful, beautiful, or even both!
Have ideas for improving it? Please let me know.