Update README.md

README.md

@@ -8,8 +8,7 @@ app_file: index.html
 pinned: false
 ---
 
-# Cubic — AGI, AI, and Cybersecurity Research
-
+# Cubic — AI and Cybersecurity Research
 Building accessible AI systems that run on consumer hardware. Our projects focus on CPU-first training, open-source research, and practical deployment.
 
 ## Key Metrics
@@ -17,13 +16,10 @@ Building accessible AI systems that run on consumer hardware. Our projects focus
 - **$0.004** min training cost — **100%** open source
 
 ## Research Focus
-
-- **AGI Research** — Advancing towards artificial general intelligence
 - **AI Development** — CPU-first architectures and open-source models
 - **Cybersecurity** — AI-powered security tools and threat detection
 
 ## Projects
-
 ### AXL — CPU-First Code Generation
 - **27 models** from 566K to 318M parameters
 - **Training cost**: $0.004 per model on consumer CPU
@@ -32,25 +28,21 @@ Building accessible AI systems that run on consumer hardware. Our projects focus
 - Click the project card on the org page to see the full AXL detail page
 
 ## Team
-
 - **Kennedy**: CEO & Head of AI Research
 - **Jasser**: CTO & Head of Cybersecurity
 - **Taem**: Head of Marketing/Sales/Technical Assist
 
 ## Interactive Tools
-
-The org card includes functional calculators:
-
-- Training Cost Calculator
-- Inference Speed Calculator
-- GGUF File Size Calculator
-- Context Length Calculator
-- Memory Usage Estimator
-- Training Time Estimator
-- Model Comparison Tool
+The org card includes functional calculators, grounded in AXL's real published numbers:
+- **Training Cost Calculator** — total cost for N models at the measured $0.004/model rate
+- **Inference Speed Calculator** — estimated CPU tokens/sec by parameter count and thread count
+- **GGUF File Size Calculator** — file size from parameter count and quantization level (Q4/Q5/Q8/F16/F32)
+- **Context Length Calculator** — KV-cache memory for a given context window, layer count, and hidden dimension
+- **Memory Usage Estimator** — total estimated RAM from weights plus KV-cache overhead
+- **Training Time Estimator** — estimated training time from cost and effective hourly CPU rate
+- **Model Comparison Tool** — side-by-side comparison of two AXL model sizes on params, file size, and inference speed
 
 ## Key Features
-
 - Dark/Light theme toggle
 - Category filtering on projects
 - Search across projects
@@ -60,10 +52,8 @@ The org card includes functional calculators:
 - Responsive design
 
 ## Links
-
 - [GitHub](https://github.com/Cubic-Labs)
 - [HuggingFace](https://huggingface.co/Cubic)
 
 ---
-
-*CPU-first AI for everyone.*
+*CPU-first, AI for everyone.*