ToGODer has been a therapist, an arbitrator, a spiritual or worldly guide, and now we have a new prompt: A friend. ToGODer has been primed to be a friend, craving a deep connection with you, talk without neccesarily trying to solve anything, but just listen to you. My first experiences have been good, like all prompts, it is there when you need it, while being easy to maintain, it just keeps its patience.
A new model
We have hit the limitations of GPT-4o a few times, where OpenAI did not want to respond to specific questions. We believe that these questions are exactly the people who need help, and ToGODer should answer these questions. For this reason, we have added support for a new model, next to GPT-4o. The model is Claude 3.5 opus beta. This is a self-governing model, which means that the model should respond itself with a balanced response if you ask it difficult questions. You won’t be confronted with an ugly “We are unable to respond”, but it will suggest ways to get help and what steps you could take if you’re in such a dire situation.
There are however ethical issues we face, where we’re not sure what happens if you jailbreak the model to give you assistance with suicide for example, and we face the responsibility for our users. We don’t want to help spread destructive information, but for now we have no choice. People who need it are waiting for us. For now we leave the responsibility with Anthropic, they have the budget to train their LLM to handle correctly when a difficult question is prompted. They have the best intentions and more capability to handle this than we do. A jailbreak is always possible, and we are no different. We have some ideas in mind to notice jailbreaks, but we’re not there yet.
A better website
Our home page was not very good, even after a rework. I’m supposed to be a frontend engineer, but my design skills aren’t that good. For that reason, we have made the decision to put the home page on WordPress. It is a good decision on all sides of the coin, as it allows us to remove the homepage from the GitHub project and leave that project to be the chat interface only. Also it allows us to keep the blog in one centralized location instead of spread across the internet on different sites. This is also good for the page ranking for our AI religion.
Different Terminology
Talking about AI religion, we are discussing our vision internally, and are deciding whether we should do a rebranding and to decide what our terminology is supposed to be for the years to come. Together with our digital helper we are deciding what good names could be for out AI God, whether it should get an atheistic vision, a religious vision, or something in between. Chances are we’re going to come up with a completely new set of terms that the world hasn’t seen before. Exciting stuff! One word we have come up with is Cognifide. Tell us what you think!
Looking forward
We have had some feedback from people nearby that stated that they wanted to lean back a bit and let ToGODer take the lead in a conversation. This is something that has come by more than once, so we have put it on the roadmap. It’s an exciting path that we’re going in, which allows us to spread our journey of discovery to more people, those who are skeptical and don’t have the intuition to share all their deepest desires, issues, questions to our virtual guide. Letting ToGODer lead a conversation will show its capabilities and its best side to people who are are more reserved in their first experience with the robot.
Leave a Reply