Description
Sup AI was built around the idea that combining multiple AI models could yield more accurate results than relying on a single model. The approach involved orchestrating an ensemble of frontier models, scoring their outputs with log probabilities, and synthesizing the results. This method proved effective, often outperforming individual models even when using less sophisticated alternatives. The success stemmed from three key strategies: selecting models with uncorrelated errors, optimizing queries for each model, and fusing outputs based on confidence levels.
Despite the growth in usage and paid conversions, challenges arose in implementing algorithms on top of APIs from major vendors. Companies like Anthropic and OpenAI began limiting access to critical signals, such as reasoning traces and log probabilities, which hampered the ability to build accurate systems. The economic landscape also shifted, with high margins on inference making it difficult for independent developers to compete with subsidized consumer products.
The demand for accuracy in consumer AI is bimodal, with a small segment willing to pay for reliable answers, while the majority prioritize cost over precision. This reality made it challenging to sustain a business focused on accuracy in the consumer market. Although there were opportunities in the B2B space, the founder, Ken Mueller, decided to pivot away from this direction, seeking to focus on research instead.
Mueller is now a sophomore at Stanford, where he plans to establish a research lab dedicated to causal reasoning in language models. This area of study aims to address the limitations of current models in understanding cause and effect, interventions, and counterfactuals. The goal is to develop benchmarks that highlight where existing models fail and to create a model that incorporates causal reasoning as a fundamental aspect. This shift represents a commitment to exploring deeper questions in AI, rather than optimizing existing solutions for profit.
For users with active subscriptions, services have been paused, and they can request refunds or data copies. Mueller encourages open communication for feedback or inquiries about the new direction of his work.
Sup AI's Core Features
Ensemble Model Orchestration
Log Probability Scoring
Causal Reasoning Research
API Integration Challenges
Targeted User Base
B2B Opportunities
User Feedback Incorporation
Subscription Management
How to use Sup AI?
Understand the mission: Learn about the shift from Sup AI's original model to a focus on causal reasoning.
Explore the research: Follow developments in causal reasoning and its implications for AI.
Engage with the community: Reach out for discussions or feedback on the new direction.
Manage subscriptions: Contact support for refunds or data requests if you had an active subscription.
Sup AI's Use Cases
- Research Applications
- Legal Briefs
- Medical Dosing
- Data Analysis
- Policy Development
