Welcome!

Hello everyone! Welcome to the first (of hopefully many) Coditum DevBlogs! This is a new series of posts that I will be authoring to keep you all updated on what I’ve been working on.

I’ll be sharing updates on the development of coditum.io, my own AI research, and other projects I’m working on. I’ll also cap each post off with a list of papers/posts that I’ve been reading this week that I found particularly interesting.

Thank you for reading, and I hope you enjoy my first attempt at something like this. Any and all feedback is appreciated!

Coditum.io

For those of you unaware, the coditum app has found a permanent home at coditum.io. I’ve been working on a number of new features and improvements, and I’m excited to share them with you all.

Some of these features are already live in the app, while most of the others will be pushed in an update tomorrow night (1/19)

Revamped Profile Page

If you open up the profile page (now accessible from the top navbar instead of a card), you’ll notice that it looks a bit different. I’ve added a new ‘History’ section that contains a table of all of your past lessons and reports. Additionally, the ‘Experience’ bar up top increases based on the number of hours you’ve taught Remember though, it only counts hours that you have reported via the app, so be sure to always fill out your reports after each lesson!

AI-powered Report Filtering

Gone are the days of manually selecting your checkpoint from a huge dropdown. Now, a report only requires you to write a sentence or two about the lesson you just taught! A language model will then automatically extract the checkpoint in the curriculum that you taught and add further metadata to the report based on what you wrote. This feature is still in beta, and only accessible by select users, so please let me know if you encounter any issues with it.

Teacher Match & the Leads Dashboard

As part of our goal of helping our teachers find clients as easily as possible, I’m happy to announce the release of the much-anticipated Teacher Match system. As opposed to the current system, where prospective clients must browse the directory to find teachers or be referred by Steven, Teacher Match will allow YOU to proactively select the clients you want from a pool of ‘Leads’. Prospective students or their parents will fill out a form describing what they are looking for, and that data will be sent to the Leads Dashboard. As the teacher, you now can claim leads from the dashboard and reach out to them to schedule a lesson. Each lead may only be claimed by 3 teachers, however, so be sure to check the dashboard frequently! This feature will be rolling out over the next week, with a full release expected by the end of the month which will include pre-made emails that can automatically be sent to clients after you have claimed them.

A Peek into the Future

I’ve been working a bit on a new update to the Copilot Chat in the app. As a reminder, it gives you all free access to GPT-4, at our expense. I encourage you to take advantage of it.

An overhaul to the existing chat system will come in the next week or two. This will include things like a better UI, chat history, model selector/settings, and more. I also plan to include custom RAG setups on top of the curriculum docs, but that probably will not be available until a bit later.

I’ve also been working a TON on what will eventually become the full Coditum Copilot. Still not sure what it will actually be called, but I’m leaning towards something like ‘Ambient Copilot’ or ‘Active Copilot’. Right now, it is being developed under the name ‘Charon’, and I can’t wait to share it with you all. I have nothing to show this week in terms of AI Research, which in large part is contributing towards Charon, but I’ve made some pretty huge advancements over the past few weeks that I can hopefully showcase soon.

What I’ve Been Reading

• – AlphaGeometry: An Olympiad-level AI system for geometry: Link
• – State-of-the-art Code Generation with AlphaCodium – From Prompt Engineering to Flow Engineering: Link
• – On the Planning Abilities of Large Language Models: A Critical Investigation: Link
• – Understanding and Coding Self-Attention, Multi-Head Attention, Cross-Attention, and Causal-Attention in LLMs: Link
• – AMIE: A research AI system for diagnostic medical reasoning and conversations: Link
• – Mixtral of Experts: Link

Conclusion

Thanks for reading! Expect to hear from me again soon 🙂

~ Ryan Heaton, CTO