I would love to have a group call. That would be amazing. I have a cal.com meeting link with my availability here: cal.com/nick-arguflow/1-hour-meeting for whenever works best for those who would want to join.

There's many already advanced professional tools such as 100 DeepAI Debater https://deepai.org/chat/debate 1st_place_medal. Debaters can fine-tune their own practice AI debater by using ChatGPT with a simple prompt
https://chat.openai.com/

I think that what we have implemented is a lot more advanced than DeepAI Debater or just plugging a prompt into the official ChatGPT UI. We have feature parity with ChatGPT and are not just an input box that doesn't save your practice debates across devices and sessions. In addition to ChatGPT, we worked our UI to place the feedback side by side with the practice counterargument so it's easier to digest, set the parameters on our API requests to get better results, engineered prompts for judging, summarizing, etc, engineered a better prompt than what you suggested for ChatGPT, and more. I am terrible at documenting work, but really need to to add a section to our README and maybe a blog post going through all of the features. We worked closely with the excellent folks at MyDebateCoach to build something that actually does offer value over using the regular ChatGPT UI or DeepAI's teaser UI.

The main weakness of the AI Debate Coach is its inability to reference quality evidence, but we are attempting to deal with that with vault.arguflow.com. The vault uses Qdrant's vector store and OpenAI embeddings to build an open and semantically searchable evidence store. Once, we have enough evidence in the vault, we plan to use a LangChain-like framework to chain requests from the AI Debate coach such that it can pull evidence from the vault and be even higher quality with its feedback and counterarguments to do do things like cite and suggest evidence. I think there's great room to chain in requests to CX_DB8 to cut the cited/suggested evidence.

Our Article Summarizer was mostly just a toy example product to play with streaming on the Edge through Vercel's Edge functions. However, it's been consistently getting the largest number of users out of what we have built thus far which is why I mentioned it. What makes it somewhat unique is that we support sites which require JS to load by using puppeteer to pull their content instead of a plain GET request.

That is why Microsoft released Cognitive Search which does ChatGPT across a large set of files and documents, and there are many other such tools and hacks coming: https://www.youtube.com/watch?v=bZQun8Y4L2A

So the problem with searching files and documents is cutting them into embeddable sections to get stored and queried, We have tried to do a "chat with anything from crates.io" project and "chat with your brief." However, it's just too difficult to parse out the best embeddable sections to make applications like that work well. I honestly think that vault.arguflow.com is a really good attempt at putting together an open embedding-based evidence store than can be queried via chained LLM vector-similarity queries. Especially once we we finish adding the social "earn karma"-like feature to it over the weekend such that there is some incentive to contribute to it beyond just supporting an "open evidence" initiative.

There's many already advanced professional tools such as 100 DeepAI Debater https://deepai.org/chat/debate 1st_place_medal

Debaters can fine-tune their own practice AI debater by using ChatGPT with a simple prompt https://chat.openai.com

I just wanted to share a side by side of what you get with Arguflow AI coach vs. DeepAI. I think the quality difference is night and day and helps to demonstrate all of the work we put in. One thing I didn't even realize until seeing the screenshots is that we put in the work to have the auto-resizing textarea and DeepAI didn't even do that. Not to mention stop streaming, regenerate, edit, etc.

image
image

doing a custom google search, for the Vault.

I didn't catch this on my first reply in-thread, but the Vault has absolutely nothing to do with Google. The vault only queries embeddings stored on our openly available Qdrant db. To add embeddings that can be queried, you just have to make an account and fill out the form at vault.arguflow.com/create.

image

Sorry about the account requirement, but we felt there was a need to be able to ban users who upload bad-faith evidence.