Home▸AI & Machine Learning▸Frontier Models & Scaling Laws: Chinchilla Compute-Optimal Explorer

📈 Frontier Models & Scaling Laws: Chinchilla Compute-Optimal Explorer

AI & Machine Learning3DAdvanced60 FPS

Details

CategoryAI & Machine Learning

RenderingThree.js · WebGL

DifficultyAdvanced

Target FPS60

ControlsModel-size slider (0.1B-1000B params) and training-tokens slider (1B-10T tokens) move a marker across the 3D Chinchilla loss surface; Optimal-frontier toggle shows/hides the compute-optimal ridge line; Snap-to-compute-optimal button re-places the current model onto the frontier at its own compute budget.

Three.jsAILLMscaling lawsChinchillacompute-optimalfrontier models

Related simulations

Agentic Workflow SimulatorWatch a Planner-Executor-Critic loop route task tokens through tools, retry on rejection,…

Guardrails and Safety Monitors SimulatorWatch AI requests stream through layered guardrails — allow/deny lists, content filters, PII…

Long-Context Retrieval SimulatorWatch chunked documents live in embedding space, queries pull the top-K nearest chunks by… ⊞ All 880 in AI & Machine Learning →

⚙ Under the hood

Explore the Chinchilla scaling law in 3D: watch loss L(N,D) form a surface over model size and training-token count, trace the compute-optimal frontier, and see whether a model is under-trained or over-trained for its compute budget — the same math behind sizing GPT-4/Gemini Ultra-class frontier models and mixture-of-experts training.

Three.jsAILLMscaling lawsChinchillacompute-optimalfrontier models

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install