SmartGPT UI Roadmap

[ymetz]

This issue is meant to be updated as the list of changes is not exhaustive

Hello everyone,

we have recently launched the SmartGPT UI as a small exclusive preview to AI Insiders. This was intended to get some initial feedback, as well as testing the online deployment without overwhelming our initial capacity.

Already, i think that SmartGPT showcases an exiting way to unluck the potential of even the current generation of large language models. Utilizing their capabilities to generate diverse output, and crucially to reflect on their generation, helps to generate higher quality output. And more importantly, SmartGPT prompting can, as we have already highlighted , enable LLMs to solve tasks that they could not solve with standard forward text generation.

But we are not done with SmartGPT. We have plans for further expansion.

Features

Right now, SmartGPT UI already has some great features:

• Generate smart prompts for your requests utilizing multiple requests, researchers and resolver prompt
• Ability to utilize multiple models, importantly using different models for original requests and researcher/resolver prompt
• Optimized default prompts (system, assistant, researcher prompts,...), but also to ability to use your own prompts
• Save and load prompt configuration
• A history of conversations, including the option to import and export conversations
• An easy way to have conversation with models of different providers via the normal chat interface

Roadmap

We are working on some improvements and even major new features. And we would love to see your support.

• Improved error messages (right now the rror messages are often empty)
• Better formatting of the output (e.g. coloring for different models)
• Groq Integration done (Add Groq API support and integration #15 - icarecti)
• Integration of additional models (Gemini API planned)
• Using different models for the initial model calls
• Enable propper chatting after the SmartGPT phase, i.e. continue the conversation after initial prompts
• Multi-Modality: Read Images as input, potentially even generate images as intermediate outputs and use similar research, resolver
• Document reading/ File upload
• Variable reasoning graphs, i.e. Smart-GPT uses a fixed structure of initial calls, researcher and resolver, maybe more flexible structures (similar to Graph of Thoughts can give improved results in some cases)
• Automated prompt tuning

and potentially more...