Sedona Thesis

Deep learning’s evolution can be framed as two distinct eras, each marking profound shifts in capability and impact. The first era spanned roughly ~20 years and focused on the development of new architectures, algorithms, and models. First era papers that introduced new architectures like AlexNet and Transformers were foundational, ultimately paving the way for today’s second era leading models like GPT-1/2/3, o1/3/4 . The exponential success of each model in the GPT and O series by OpenAI has proven the validity of the scaling hypothesis and opened the door to a new focus - creating novel problems for advanced models to learn and solve

This new paradigm as the eras transition from scaling up pretraining to scaling RL behind closed doors exacerbates a new challenge for small and large businesses and enterprises alike: "adopt AI or die". Adopting AI is no longer enough because in the RL era the intelligence is naturally "spiky": meaning capabilities are not distributed. For example, ChatGPT and Claude are excellent at Python--possibly because their own codebases are written in Python--but struggle with writing more functional languages like Rust and Haskell. Overspecialization is not only a problem unique to programming it also affects other disciplines and workflows as well. A classic example is LLMs are excellent at customer service but due to context rot they struggle recalling the product or service they are providing customer service for causing confusion for customers. Clearly, "the age of intelligence" is here but the problem for businesses who are trying and failing to leverage AI is it is not fully distributed yet.

Our solution to developing AI capable of automating work is Sedona. Sedona is a marketplace and ecosystem focused around a fullstack suite of tools to train and customize models in a cost-effective manner and requiring no prerequisites in deep learning. Our core belief is training an LLM should be as simple as defining the problems that need to be solved and then click "train" and several hours later click "deploy".

Our core belief aligns with our vision of a better future. A future where AI is more open and is more accessible than it is today. These values are important to us because we believe strongly in the next 5-10 years nearly all work will be automated by AI agents. Therefore, we hold strongly ownership of the models cannot be as centralized as it is today. Something must change so future progress will widen opportunity instead of shrinking competition...

We believe Sedona will be an important part of the change towards a better future.