power users logo

Constitute

ChatGPT: A web platform offering chatbot assistance for productivity.
traffic icon
Monthly Traffic: Insufficient Data

What is Constitute?

Constitute AI is a company that creates products aimed at adding time to people’s lives. Their main product, Scigic, is a browser and assistant designed to be a single point of contact for all digital interactions. It allows users to access their favorite web apps using a shortcut on their computer. The company also focuses on developing a harmless AI assistant through self-improvement, which they refer to as “Constitutional AI.” This method involves both supervised learning and reinforcement learning phases, with the goal of creating an AI assistant that engages with harmful queries by explaining its objections to them.

 


 

⚡Top 5 Constitute AI Features:

1. Constitutional AI: A method for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. It involves both a supervised learning and a reinforcement learning phase, with the process involving sampling from an initial model, generating self-critiques and revisions, and then finetuning the original model on revised responses. In the RL phase, samples from the finetuned model are used to evaluate which of the two samples is better, and then a preference model is trained from this dataset of AI preferences. The preference model is then used to train with RL using it as the reward signal, i.e., ‘RL from AI Feedback’ (RLAIF). This method allows for the training of a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them. Both the SL and RL methods can leverage chain-of-thought style reasoning to improve the human-judged performance and transparency of AI decision making.

2. Harmlessness from AI Feedback: A process that makes it possible to control AI behavior more precisely and with far fewer human labels. It involves training a preference model from a dataset of AI preferences, which is then used to train with RL using the preference model as the reward signal. This method allows for the training of a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them.

3. Supervised Learning: A phase in the Constitutional AI method where samples are taken from an initial model, then self-critiques and revisions are generated, and the original model is finetuned on revised responses. This phase helps to improve the performance of the AI assistant by allowing it to learn from its own mistakes and revisions.

4. Reinforcement Learning: A phase in the Constitutional AI method where samples are taken from the finetuned model, and a model is used to evaluate which of the two samples is better. The preference model is then trained from this dataset of AI preferences, which is then used to train with RL using the preference model as the reward signal. This phase helps to further refine the AI assistant’s behavior and decision making.

5. Chain-of-thought style reasoning: A method used in both the SL and RL phases of the Constitutional AI method to improve the human-judged performance and transparency of AI decision making. This method allows the AI assistant to provide more detailed and contextually relevant explanations for its decisions, making it easier for users to understand and trust the AI’s recommendations.

 


 

⚡Top 5 Constitute AI Use Cases:

1. Supervising other AIs: Constitutional AI can be used to enlist the help of AI systems in supervising other AIs. This can be particularly useful in situations where human oversight is limited or not feasible, such as in large-scale AI systems or in situations where human intervention could be harmful or inefficient.

2. Training harmless AI assistants: The Constitutional AI method can be used to train AI assistants that are harmless and non-evasive, even when engaging with harmful queries. This can be particularly useful in situations where the AI assistant needs to provide explanations for its objections to harmful queries, such as in customer service or support settings.

3. Improving AI decision making transparency: The Constitutional AI method can be used to improve the transparency of AI decision making by providing detailed and contextually relevant explanations for the AI’s decisions. This can help users to better understand and trust the AI’s recommendations, leading to more effective and efficient use of AI systems.

4. Controlling AI behavior more precisely: The Constitutional AI method allows for the training of AI assistants that can be controlled more precisely, with far fewer human labels required. This can be particularly useful in situations where the AI assistant needs to be trained to behave in a specific way, such as in customer service or support settings.

5. Leveraging chain-of-thought style reasoning: The Constitutional AI method can be used to leverage chain-of-thought style reasoning to improve the human-judged performance of AI decision making. This can be particularly useful in situations where the AI assistant needs to provide detailed and contextually relevant explanations for its decisions, such as in customer service or support settings.

Share:

View Constitute Alternatives:

Login to start saving tools!