Not Diamond isn’t like other chatbots you’ve used. Not Diamond automatically calls the best model for any prompt and improves in real-time based on your feedback, continuously learning your preferences. Not Diamond is the last chatbot you’ll ever need.
Hey Product Hunt!
I’m Tomás, co-founder and CEO of Not Diamond: chat.notdiamond.ai
Not Diamond is the first AI chatbot that automatically recommends the best AI model for every message and improves in real-time based on your feedback, continuously learning your preferences for when to use which model.
The future is multi-model. Every week (often every day), a new state of the art AI model is released. These models vary widely in their abilities across domains, and over the past year I found myself constantly jumping back and forth between chatbots, manually comparing different models to get the best answer. More complicated still, my preferences kept evolving over time as the models themselves did.
We built Not Diamond because we want to take the guesswork out of AI. Now you can leverage the best AI models with the keen-eyed discrimination of a veteran AI engineer. Even better, Not Diamond learns your preferences for which model to use on everything from conversation to coding to creative writing.
This is an incredibly powerful tool packed with features—here are a few of my favorites:
🍢 Automatic AI model routing: Not Diamond automatically chooses the best LLM for every message you send, with new models added every week.
💫 Real-time learning: Unlike traditional chatbots, Not Diamond learns from your feedback in real-time, ensuring that your experience is always getting better.
⚔️ Arena mode: Turn on arena mode to compare models head-to-head in a blind selection process, supercharging your personalized model recommendations.
🔥 SOTA quality: Not Diamond beats every individual LLM on every major benchmark and continuously improves as new models are released.
🛹 API access: Not Diamond is also available via our API so that you can implement automatic AI model routing in your own AI applications.
🪩 Open-source: Our chat app is based on an incredible open-source template built with the Not Diamond API by McKay Wrigley.
🌸 Free to use: Not Diamond is entirely free to use, with no limits on how many messages you send to our chat app.
I’m super excited to share this with you all—I’ve been using it nonstop and it’s a game changer. Try it out and lmk what you think!
chat.notdiamond.ai
In a way this is like the MoE architecture, but at a higher level of abstraction.
Congrats on your app release, @tomashernandokofman! The app runs like a charm, and the navigation is really straightforward. The UI design is clean and the color scheme is very inviting.
Congrats on the launch! I'm curious to find out how would this generate revenue, what is the plan like?
model: Perplexity
latency: 1175ms
cost: $0.0027
Really liked the feature showing the per chat cost up-front, it will become very easy for developers to benchmark and understand cost vs accuracy tradeoffs.
Congratulations @tomashernandokofman, @tzeyang and @jakiki for beating the SOTA on every benchmark. Keep up the amazing work! 🚀
This is fantastic! Have you considered adding a feature that allows users to see a history of model recommendations and their performance?
Just looking for a chatbot also another tool falls into the list to test ;)
Congratulations on the launch!
Congrats on the launch! @tomashernandokofman
Have a few questions on your beautiful looking product:
1) How many AI frameworks do you use for Not Diamond? I can only think of 3 or 4 popular ones (i.e. ChatGPT, Claude).
2) In your opinion what's the trade-off between using a lot of specialized AI frameworks specialized in certain fields and having to deal with routing and switching cost vs. using one generalist AI framework?
3) Can you quantify how much better a particular AI framework is in a certain field vs. another AI framework? (i.e. ChatGPT is 50% more accurate than Claude in coding)
Again, would love to learn more about your work and your product - thanks for making this!
I just tried Not Diamond and it works pretty well. Does it only support txt and pdf documents? I saw some paid large language models. Can we use this product for free?
I love the idea of automatically routing queries to the optimal AI is amazing. Good one team😁
The NOT Diamond chatbot AI offers a unique conversational experience that prioritizes user engagement with a blend of natural language understanding and context-aware responses. Its advanced algorithms enable seamless interactions, making it a versatile tool for customer support, personal assistance, and casual conversation. With continuous learning and updates, NOT Diamond chatbot AI stands out as an innovative solution in the realm of artificial intelligence-driven communication.
Wow! It would be awesome to collaborate or even integrate your chatbot with Boltathread!
🎉 Congratulations! 🎉
Just out of curiosity, I was wondering about the ai-model-router. It seems to be a classifier that categorizes conversations based on their semantics and then forwards them to different specialized models for responses.
I'm curious if using a small model in your own application to classify queries into limited categories before submitting the results to the corresponding specialized models would achieve similar effects. Let's set aside the potential delay caused by adding this extra layer of classification for now.
Here's a prompt example for reference:
```
You are a professional assistant, and your task is to correctly understand user requests and classify them. You are only allowed to search within the given range to determine if a corresponding skill needs to be executed. If the skill is not within the given range, return your result directly; otherwise, provide the name of the required skill. Always follow the format I provide.
## Available Skills
1. email_writer: I can send email to given users. required email address and email content.
2. weather_forecast: I can search for the weather of a given location. Require location.
3. home_assistant: I can help query the sensor data or send control commands to your smart home system. require specific commands like: open the light in the kitchen.
## Response Format
{
skill: write_email, // email_writer, weather_forecast, home_assistant, NO
confidence: 0~1, // give the confident score of this classification
}
```
What do you think? Would this approach work? 🤔
Interesting idea - I also like that it's open source!Give it a try and see how it works
It's cool. User will not be overwhelmed when new models are releases everyday. They don't need to care about the fact that they can miss out new suitable models for their purpose. You're taking out the task of selecting models for users, which they likely might not know which is best for them.
I also love the concept of keeping and passing on the contextual conversation across differen model.
It's coold that everything is free now but wonder how you can sustain it given the cost to querry from other models are costly and limited?
btw, congratsss!
