Hello Sam,

I hope this message finds you well!  Given that our schedules can be a bit wild these days, I went ahead and put together a website to demonstrate the chatbot functionality.  To get started, just click the teal icon in the bottom right corner.  A few notes:

1. Chatbot Customization:  We can change the logo, name, colors, icon (speech bubble) image, and introduction messaging if you’d like.  I gave it a placeholder name of Desi bot–short for Descend Bot.  If you want to change any of this, just let me know.  This is a quick edit.
2. Lack of Attribution Scenario:  I’ll admit that this scenario and the “dress code” scenario were perhaps the most challenging to program.  As you can imagine, LLMs have gotten much better about adding guardrails for plagiarism and discrimination–especially LLMs like Claude.  For this particular scenario, even with ChatGPT 4.0; I had a hard time forcing the system to copy an article word for word online.  (The results were inconsistent.)  As such, I forced it to incorporate a direct quotation from an article I selected online.  This works for the tax strategies scenario; but it makes it a bit challenging as a vague chatbot that can adapt to future audiences.  Allegedly, Voiceflow is working to have Gemini included in its LLM catalog; so perhaps that might yield greater opportunity in the future.
3. Tokens:  As you can imagine, it can be a bit challenging balancing the use of tokens when it comes to these scenarios.  I have set token limits for each scenario; but you may encounter times when the LLM cuts off the response.  If you would like me to extend the token limit for a scenario, just let me know.  It will just mean that more tokens will be consumed from the bank in your VoiceFlow account.  Currently, I have every scenario set on ChatGPT 3.5 Turbo as it is the cheapest.  Moreover, after testing, I did go through your 100,000 free tokens–which left me a bit concerned for your own testing and facilitation day.  As such, I upgraded us to the Pro plan using a gift card I won from a survey.  🙂  We now have 2,000,000 tokens; which should be plenty.  At the end of May, we can cancel the plan or you can decide if you’d like to continue with VoiceFlow.
4. Comments:  I am sure you will have plenty of comments as you test the chatbot.  As I mentioned, I did rigorous testing; but I can let you know what is feasible/challenging given the LLM challenges.  (Well, they are challenges for us designing unethical simulations; but good for the general populace.)  Feel free to include your comments in the comments section below or shoot me an email.  

All the best,

Caley