Rushing Headfirst into a Sci-Fi Future Filled With AI Agents
The Bet
On June 4 2024 I met a bet for $1000 that we would have AGI within 2 years.
I started this project to help speed things along.
The bet imposes no conditions on the class of models or architectures that will lead to AGI.
All projects listed below are in service of this goal.
"AGI in 2 Years."
- The Agent Guy, June 4 2024
Thesis
This is my current thesis regarding AGI and AI Agents. This is subject to change.
First, Agents will be the key to unlocking AGI. There is more to intelligence than neurons firing. For humans, having tools and varied output modalities played a huge role achieving our current status as the dominant intelligence.
Vision is the most important modality. The work done here will rely on pure vision as much as possible. Language is important for reasoning, but when it comes to I/O with digital (or even physical) interfaces, vision generalizes better than anything else.
On open sourcing; no real reason for it. I simply see no reason not to.
The first project will be a general-purpose vision-based Android Agent.
Humans spend a lot of time and get a lot of work done on smartphones. Additionally, Android gives us constrained and well-defined action and state spaces in a robust sandboxed emulator environment easy for us to control programmatically.
Deployment as an app is possible but is not an immediate concern. Testing will be carried out using adb.
