Experiments run on the SVHN dataset.
AI & ML interests
Deep learning, research
Recent Activity
Organization Card
"Wonder. Ask. Answer. Repeat."
Dawntasy is an AI R&D company focused on applying cutting-edge research in artificial intelligence to solve real-world problems with both efficiency and effectiveness at the forefront.
Key areas we do:
- AI safety, AI interpretability, adversarial attacks
- AI agents
- theoretical learning theory
- philosophy of AI
- natural language processing, transformers, state-space models
- training phenomena, grokking, double descent
- continual learning, catastrophic forgetting
- regularization theory, model control and AI alignment
Key products we are developing:
- Whala Agent. A general-purpose autonomous AI agent designed to automate workflows efficiently and intelligently in a workspace. Supports HITL and HOTL safeguards.
- Whala QA-QC Agent (Research Preview). An fine-tuned AI agent tailored toward QA/QC automation. Exhaustively checks products for pre-production risks before launch.
datasets 0
None public yet