Were designing a tiny autonomous research agent
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
Developers are designing a tiny autonomous research agent, initially a 43m parameter model trained on 3m arXiv abstracts. The goal is to create a local agent capable of autonomously conducting research by searching the web and documenting findings. They are experimenting to see if a language model trained exclusively on science, physics, and technology can make new discoveries. The team is seeking contributions of fine-tuning datasets and suggestions to help shape the agent's behavior and achieve its objectives.
This base model is only 43m parameters trained on 3m arXiv abstracts. We plan to continue pre-training and post training. If you create fine-tuning datasets or if you know of any datasets that can help shape the behavior for our goal we appreciate all contributors. The goal is to make a local agent that can autonomously do research. Its just a simple loop to search the web & document its findings as an experiment to see what is possible. If we train a language model on nothing but science, physics and technology can it make new discoveries?
We are testing this by creating fine-tuning examples that contain a pattern of asking questions and answering them until coming to a conclusion from first principles. If you have any suggestions to achieve the goal we are all ears. Please leave a comment.