I posted when I had more operations including unary, binary, and matmul but got a lot of help from Claude when writing that and realized I didn’t really understand broadcast operations so got rid of all that and started fresh.
2. It’s clear you didn’t read the rules folder because it’s a rule to tell cursor to teach me
3. You say you don’t understand why TensorData can be a single integer. Scalars are tensors too. I’m trying to support PyTorch’s spec and ops as much as possible which is why I’m supporting it. The obvious use cases are for reduction ops and scalar unary ops.
Right now it's just a tensor manipulation lib but will be adding an autograd engine soon. It's been fun learning about strides and doing matmuls by hand and then coding it without numpy.
Thanks for sharing! I used to teach students to build ML algorithms from scratch (everything from Markov chains to multilayer perceptrons and convolution neural networks) - I rewrote some of my notes in TypeScript here:
Educators and Parents can use AI to generate Interactive Activities and Custom Analytics so that you see how students think and not just what they know. You also get recommendations to follow up.
Froggy can make anything from a simple alphabet matching game to an electronic circuit building activity.
Traditional learning analytics treats students as numbers: 67%, B+, etc. You're also stuck with activities that was not made for your kids, their interests and needs.
When students view education as a transaction, they'll use the best tool to complete it. I don't think AI policing or banning will solve this.
My hope is that when you show them that their learning journey including the struggles, hesitations, moments of progress, and successes are cared for, they would want to put in effort. They need to know that people care about them.
Froggy is a coding agent that generates activities, tracks student events, analyzes the events, and then generates a dashboard personalized for every student and the whole class.
You can revise your activities, analyze student data from multiple perspectives and learning science theories, and generate as many dashboards as you want! This was inspired by the Education Report shared by Drew Bent from Anthropic where we see that Educators are already using AI to generate interactive activities for STEM but we need detailed and personalized analytics to close the feedback loop.
I made this with Claude Code (Sonnet 4.5) in 2 days so its not perfect but it was easy thanks to the
ai-sdk, shadcn and ai-elements. This was the most efficient I’ve been with Claude Code because in the beginning I was reviewing every single edit to setup a good foundation, types, and structure for it to learn from. Highly recommend starting with Data models, DB schemas and types before making your agent write code! I got into this habit after reading the Data Oriented Programming book and its been very helpful. Planning ahead in Notion gave Claude Code more context about the purpose of the app and my initial DB schemas.
I know that teachers and parents are already busy so this platform would have a gallery of activities to choose from and customize if needed just like v0 and Scratch. Would love to get feedback and thoughts!
I've seen a lot of purple gradient websites made by LLMs and I was curious if we can change a model's favorite color by messing with the activations instead of prompts.
So I used Representation Engineering with just 1 pair of contrastive prompts to make Mistral-7B prefer orange as its favorite color.
I used the repeng library by Theia to test this out.
Next I'm gonna implement it from scratch to understand why this even works.
I wonder if we can introduce "taste" into a model with methods like this.
The paper is "REPRESENTATION ENGINEERING:
A TOP-DOWN APPROACH TO AI TRANSPARENCY".
Will post again after making more progress.