GPT-plugins, signs of AGI, and an Open Letter – Live and Learn #14
Welcome to this edition of Live and Learn. With AI development picking up speed, it feels like new things are happening every day. I sometimes think that writing this newsletter only once every two weeks is not enough to catch up with all the awesome stuff that is happening. This edition will focus on AGI and the discussion that has ensued after ChatGPT Plugins and GPT-4 have been shown to the public.
✨ Quote ✨
Sometimes it's weird to remember that we're all effectively competing to hit the right keys on our keyboards in the right order and that if we do it for long enough we can buy a house.
– Nik Milanovic – (source)
🖇️ Links 🖇️
Open Letter by the Future of Life Institute. There are a lot of people who want to stop further progress on bigger and more powerful AI models for at least 6 months. They want to do that so that we have enough time to learn and understand how these models work and why they generalize so well. They argue that we need more understanding to judge the safety of these models and not accidentally produce a genie that can't be put back into the bottle. They claim the point where AGI is close enough that things can quickly become an existential risk is now. Therefore we should proceed with the utmost caution.
Sparks of Artificial General Intelligence by Microsoft Research. This is a paper that is well worth skimming at least (the whole thing is 155 pages long). Here's a short excerpt from the abstract: "Given the breadth and depth of GPT-4's capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system." I think this shows that the people who signed the Future of Life Institute's Open Letter are right to worry about even larger models. Larger models make unexpected jumps in thinking and reasoning capabilities, and maybe there is a threshold after which "true" AGI emerges all of a sudden.
ChatGPT Plugins by Open AI. In a nutshell, OpenAI just released a way to give their GPT-4 model superpowers. Now, with access to real-world integrations, ChatGPT can not only tell you how to do tasks but do them on its own. This gives me the creeps. To think about an intelligent algorithm, that can write source code, execute it, debug it, browse the internet, and do Zapier integrations borders on insanity. For now, the set of plugins is very limited, but in the future, any work on a computer can be completely automated, as long as you can describe it in plain English.
Attention is all You Need by Packy McCormick. This post details how ChatGPT plugins are a building block toward the perfect business model of OpenAI. ChatGPT already is a central place people go to for doing work more effectively. Because of that, it has the attention of a lot of people. This attention makes ChatGPT a useful marketing platform for companies searching for new customers. Eventually, integrating with ChatGPT becomes a necessity because that's where most of their potential users are. The companies that do offer ChatGPT integration will survive, those that don't won’t. And the plugins these companies design and build will in turn further accelerate the usefulness of ChatGPT. This in turn will grow its user base even more, which will increase the value ChatGPT integrations will have for companies and so on. Until ChatGPT eventually becomes the ultimate tool for everything you would ever need or want to do on a computer.
AutoGPT by Torantulino. The logical continuation of the HustleGPT idea from a tweet I shared in the last newsletter. This is a GitHub repo that is trying to fully automate business generation and execution with GPT-4. The idea is to let GPT-4 prompt itself over and over again and give it the possibility to inspect and run its code and ideas in Python. It's still broken when used in complex scenarios, but even the thought of this ever working is uhm... interesting to say the least.
AI video generation - Gen2 by Runway. This demo is insane. We will soon have full-feature films being created by a single person and a tool like this. WTF. Even creative jobs like those of movie directors and actors will change dramatically in the future. Think about a "Will Smith" or "Jennifer Lawrence" AI you could buy so that your scenes are "acted" out by these digital copies. And then maybe even tell the AI that the film should look like it was directed by Steven Spielberg. And then you get a crazy world, where movies can be created by everybody. A world where ChatGPT has automated everything happening on computers, while GPT-5 is being researched as the starting point for the singularity. The next few years will be a fun rollercoaster ride toward AGI.
Solugen by Elliot Hershberg. This blog post introduces the biotech company Solugen. It pales a little bit in comparison to the craziness of the AI world but I still wanted to include it here. Solugen is a crazy innovation hub, building the future of chemical manufacturing, right now. They are using biological enzymes within a "bio factory", to produce many of the chemicals needed for our everyday products. However, they do so way cheaper and carbon-free, because they pull the carbon necessary for the chemicals out of the CO2 from the air. This is nothing short of revolutionary. It seems like magic now, but soon bio factories like the ones Solugen is building will be commonplace and manufacture everything, from clothes to rocket fuel.
🌌 Midjourney 🌌
Isn't the quality of these images just insane?
🎶 Song 🎶
One More Day by Lena Raine
I hope you found this newsletter useful, beautiful, or even both!
Have ideas for improving it? Please let me know.