We're hiring a Machine Learning research scientist or engineer to work on active learning for small-data and data-limited domains, especially biology. To do this, we're developing RL-flavored transformer-successor systems that learn a model of the world through interactive observation. These systems then mechanistically reason over that model to select actions, infer hidden states (e.g. perception), and assess uncertainty (e.g. for improving the model, counterfactual reasoning, and constraint satisfaction). Unlike other approaches leveraging LLMs, we do not pretrain with human data - competency is wholly bootstrapped. (By induction, if we can go from 0 -> 1, then we can also go 1 -> N.) This research is funded by Schmidt Futures, hence we will publish our results and encourage collaborations. Please send a message if this interests you! tim@springtail.ai
About us
Springtail AI is developing algorithms and architectures, loosely inspired by the brain, which enable active learning -- that is, learning through experimentation and observation. This is to create agents that excel in environments where there is no prior data -- and to do new things. (c.f. LLMs). As part of this, we are researching transformer-successor and hybrid architectures & algorithms that improve sample efficiency and open-ended iteration. We deeply believe that enabling broadly capable information pumping will be transformative to society, including software development (imbuing users desires into code) and scientific research (driving understanding in areas where humans are challenged, e.g. population-wide diseases).
- Website
-
https://springtail.ai/
External link for Springtail AI
- Industry
- Research Services
- Company size
- 2-10 employees
- Type
- Public Company
- Founded
- 2022
- Specialties
- Machine Learning, Program synthesis, and Biological data