To set the threshold for ASI, the goal is to continue developing this project to build an LMM powered agent capable of solving all ARC-AGI 3 tasks at $0 cost with a 100% success rate across all games.
I know this sounds like a wild goal, but I believe it is achievable with the technology we are building and the philosophy we are following. I will keep this issue open until we achieve ASI, maybe by next month. When it is closed, it will mean we have reached it. A humble goal, but, let's keep building toward this end state. I believe I can improve this framework in the next week into one that can build lmm powered agents. We already have web search at $0 cost, solving math equations using cpu only and no gpus at all.
The hardest part is enabling the agent to discover a sequence of mathematical equations and mb computer vision related steps needed to complete each game.
All PRs that add support for agent-building in this framework will be tagged to this issue.
We can do this!
To set the threshold for ASI, the goal is to continue developing this project to build an LMM powered agent capable of solving all ARC-AGI 3 tasks at $0 cost with a 100% success rate across all games.
I know this sounds like a wild goal, but I believe it is achievable with the technology we are building and the philosophy we are following. I will keep this issue open until we achieve ASI, maybe by next month. When it is closed, it will mean we have reached it. A humble goal, but, let's keep building toward this end state. I believe I can improve this framework in the next week into one that can build lmm powered agents. We already have web search at $0 cost, solving math equations using cpu only and no gpus at all.
The hardest part is enabling the agent to discover a sequence of mathematical equations and mb computer vision related steps needed to complete each game.
All PRs that add support for agent-building in this framework will be tagged to this issue.
We can do this!