Get live statistics and analysis of echo.hive's profile on X / Twitter

I built a biologically inspired spiking neural network from scratch and it learned with %5 accuracy to do addition :) There is no backpropagation, no artificial loss functions - just spikes, synapses, and dopamine-like reward signals. it uses STDP -> "Spike-Timing-Dependent Plasticity" with modulated rewards This is super fun and I will try to get it to learn with better accuracy. I also need to better understand how all the moving parts fit together Link to source code in comment which has a detailed readme and html with animations explaining how it all works

deepseek-r1 is now available in Cursor. Just go to settings>models> then click it on! 🥳 it won't work with the agent tho

I built a Continuous Attractor Network with 3 motor neurons which learn to guide the mice towards food using Reward‑modulated Hebbian plasticity No backpropagation and no SGD. In this case at first, the networked learned to circle around the food 😯 Source code is open source. link in comment 36 neurons in the CAN detect where the food is, each neuron in this network connect to every motor neuron. At first the connections are random and weak and they get stronger as mice approaches and reaches food. network does this by keeping track of which neurons fired together to achieve this and strengthen those connections If you find this interesting check out my Spiking neural networks project. I will put a link to that in comment as well

Spiking Neural Network from scratch achieves 8% accuracy. no backpropagation or SGD I created a genetic hyper parameter optimizer and it now, on average, can get 8% accuracy which is ~3% above chance Link to source code with a detailed video and markdown explanations in comment it also usually starts lower than 5% and slowly improves then eventually can start dipping below 5%, all of which leads me to believe there is glimmers of learning taking place. sometims it is stabel around 7-8-9% for a long time there is no backpropagation or SGD. it learns via STDP (spike timing dependent plasticity) and a reward mechanism each example is presented n many times (500 in this case) which produces spike train which leads to eligibility list, at the end of a turn based on if the answer was correct or not we adjust the weights using the reward as a multiplier spike timing keeps track of the sequence of neuron firings and which ones were more likely to have lead to the correct answer let me know what you think

Memory systems for LLMs are easy 🥳 here are 5 implementations as objects 1) Vector memory 2) summary memory 3) time window memory 4) keyword memory 5) hierarchical memory 1)Vector Memory System ------------------- This memory system uses OpenAI's embeddings to create vector representations of messages and enables semantic search through the conversation history. How it works: 1. Each message is converted into a high-dimensional vector (embedding) using OpenAI's embedding model 2. When retrieving memories, the query is also converted to an embedding 3. Cosine similarity is used to find the most semantically similar messages 4. Messages above the similarity threshold are returned, sorted by relevance Key features: - Semantic search capability (finds conceptually similar content, not just exact matches) - Maintains a fixed-size memory buffer (MAX_MEMORIES) - Uses similarity threshold to ensure quality matches - Returns top-k most relevant memories Use cases: - Finding semantically related previous conversations - Answering questions about past discussions - Maintaining context across long conversations --- Want to level up your Cursor game? I’ve created a 36-chapter course on mastering Cursor. Check it out via the link in my bio! featuring 22 hours of content where we build full apps from scratch in each chapter. each chapter is short and independent and designed to get your started quickly. ---

Sonnet MAX is coming to Cursor! What is it?

2 Cursor Composer agents working together on the same project. one builds the project, the other reviews once the first agent is done and writes a report. cursor rules file is in comment to achieve this. first agent is instructed to build the project second agent is instructed to write a monitoring script to determine when the project is done then to review the project for errors and write a report This prompt can be improved. This is meant as a baseline If you are interested in learning how to use Cursor more in depth, I have a 28 chapter 1000x Cursor Course building full apps from scratch with 19 hours of content. link is in my bio if you are interested

📚🤖 Open Source: AI reads PDFs Page-by-Page, extracts Knowledge & Summarizes 🥳 Repo in comment 💝 - 📚 Automated page by page PDF book analysis and knowledge extraction - 🤖 AI-powered content understanding and summarization - 📊 Interval-based progress summaries - 💾 Persistent knowledge base storage - 📝 Markdown-formatted summaries - 🎨 Color-coded terminal output for better visibility - 🔄 Resume capability with existing knowledge base - ⚙️ Configurable analysis intervals and test modes - 🚫 Smart content filtering (skips TOC, index pages, etc.) - 📂 Organized directory structure for outputs ---- Want to level up your Cursor game? I’ve created a 28-chapter course on mastering Cursor. Check it out via the link in my bio! featuring 19 hours of content where we build full apps from scratch in each chapter. each chapter is short and independent and designed to get your started quickly.

Patterns of Uncertainty node graphs with Markov chains Version 2 Markov chains are fascinating! it is a set of probabilities (transition matrix) which determines the next state of a system or a branch based on its current state. it is like saying if you are now "up" then go left 30% of the time and right %20, but for many states and many possibilities Link to code in comment. Let me know what you think

Finished "Neuroscience" lecture series 🥳 started on "Manifold learning", same teacher aka "dimensionality reduction" I always thought in ML when we say "this data is 1000s of dimensions" etc, that this is the accurate dims of representation of it and dim reduction is an approximation so our eyes can perceive it at some level But actually, the true representation of a 1000 dimensional data can be 50 dimensions perhapsl or even 3 or 2 so those 1000s Dims are just the space that represents it but that doesn't mean that is the most meaningful representation this is a quick introductory course and I highly recommend it. only 7 short lectures She really teaches well and the lecture talks about methods as well like t-SNE, UMAP, PCA etc I will put a link in comment

I built a biologically inspired spiking neural network from scratch and it learned with %5 accuracy to do addition :) There is no backpropagation, no artificial loss functions - just spikes, synapses, and dopamine-like reward signals. it uses STDP -> "Spike-Timing-Dependent Plasticity" with modulated rewards This is super fun and I will try to get it to learn with better accuracy. I also need to better understand how all the moving parts fit together Link to source code in comment which has a detailed readme and html with animations explaining how it all works

deepseek-r1 is now available in Cursor. Just go to settings>models> then click it on! 🥳 it won't work with the agent tho

Sonnet MAX is coming to Cursor! What is it?

Attempting the Impossible: Teaching a Neural Network to Predict Pi's digits Here's what we're doing, in the spirit of futility mixed with curiosity: We grabbed 10,000 digits of Pi and gave each one a unique binary "fingerprint, like assigning a bar code to every position (N choose K encoding) like 11111111100010000000,5 (4th digit wihch is 5) Then we built a ridiculously overengineered neural network (1024-neuron hidden layers, 7 residual blocks, fancy learning-rate schedules, the works) and asked it: "Here's the bar code for position 10,857. What digit comes next?" (predicting the next 1000 unseen digits) it pretty much memorizes the training data but guesses at chance (10%) for unseen digits :( although Pi's digits are mathematically proven to be randomly distributed, one can still try right? But I had to see it with my own eyes 😝 I will report back if a miracle happens

Spiking Neural Network from scratch achieves 8% accuracy. no backpropagation or SGD I created a genetic hyper parameter optimizer and it now, on average, can get 8% accuracy which is ~3% above chance Link to source code with a detailed video and markdown explanations in comment it also usually starts lower than 5% and slowly improves then eventually can start dipping below 5%, all of which leads me to believe there is glimmers of learning taking place. sometims it is stabel around 7-8-9% for a long time there is no backpropagation or SGD. it learns via STDP (spike timing dependent plasticity) and a reward mechanism each example is presented n many times (500 in this case) which produces spike train which leads to eligibility list, at the end of a turn based on if the answer was correct or not we adjust the weights using the reward as a multiplier spike timing keeps track of the sequence of neuron firings and which ones were more likely to have lead to the correct answer let me know what you think

My thought is that AI assistance accelerates coding from 1 to 100x depending on the complexity with current capabilities You can single shot a 1000 line app in a minute. That is when it is 100x But as complexity increases, Xs are reduced, eventually all the way back down to near 1x 🙂 And at maximum compexity, quite possibly, to slower than 1x if you haven’t taken the time to understand the code AI has written. Which can occur more often than we would like to admit. What do you think?

I built a Continuous Attractor Network with 3 motor neurons which learn to guide the mice towards food using Reward‑modulated Hebbian plasticity No backpropagation and no SGD. In this case at first, the networked learned to circle around the food 😯 Source code is open source. link in comment 36 neurons in the CAN detect where the food is, each neuron in this network connect to every motor neuron. At first the connections are random and weak and they get stronger as mice approaches and reaches food. network does this by keeping track of which neurons fired together to achieve this and strengthen those connections If you find this interesting check out my Spiking neural networks project. I will put a link to that in comment as well

2 Cursor Composer agents working together on the same project. one builds the project, the other reviews once the first agent is done and writes a report. cursor rules file is in comment to achieve this. first agent is instructed to build the project second agent is instructed to write a monitoring script to determine when the project is done then to review the project for errors and write a report This prompt can be improved. This is meant as a baseline If you are interested in learning how to use Cursor more in depth, I have a 28 chapter 1000x Cursor Course building full apps from scratch with 19 hours of content. link is in my bio if you are interested

I just built a script where Cursor composer uses clause computer use tool to instruct it to instruct itself to do things!!!! now we are talking! still working on it. update soon. but read below for some mind twisting to see how mind bending this is take a look at the cursor rules file and the system message for claude computer use api call: cursor rules: when you are asked to use yourself. please use the use_yourself function in use_yourself.py. this exists and you can call this directly with your request. When instructed to use yourself, your argument to the function will be the instructions to an AI agent to use you. so craft the function parameter to instruct the agent to instruct you for whatever the user is asking for your parameter should start with "I am cursor, please instruct me to..." computer use system message: * YOU ARE USING A CODE EDITOR CALLED CURSOR VERY SIMILAR TO VS CODE. PLEASE INSTRUCT THE AGENT ON THE RIGHT PANEL. you must enter your instructions in the right bottom of the screen. you will be instructing an AI agent there to do what is being asked You are not allowed to do file operations. you are only allowed to use the cursor code editor in the window that you see. insert your instructions to the agent there at the bottom right where it says edit code

OMG! I talk to Gemini 2.0 Flash with REALTIME VIDEO CHAT while testing its reasoning capabilities with my test script in Cursor 🤯 This is some serious next level stuff! 🚀

Finished "Neuroscience" lecture series 🥳 started on "Manifold learning", same teacher aka "dimensionality reduction" I always thought in ML when we say "this data is 1000s of dimensions" etc, that this is the accurate dims of representation of it and dim reduction is an approximation so our eyes can perceive it at some level But actually, the true representation of a 1000 dimensional data can be 50 dimensions perhapsl or even 3 or 2 so those 1000s Dims are just the space that represents it but that doesn't mean that is the most meaningful representation this is a quick introductory course and I highly recommend it. only 7 short lectures She really teaches well and the lecture talks about methods as well like t-SNE, UMAP, PCA etc I will put a link in comment