AI/ML●●●Banger
Watch a neural net discover molecules by arguing with itself
Two RL clones duel in-browser with epiplexity selection — no human tuning the explore-exploit knob.
WizardryRabbit HoleBig Brain
cmaicurious
1020d ago
Back to browse
178K Parameter Neural Net That Wins Poke(rogue)like
by farcaster·Jun 4, 2026·3 points·0 comments
AI Analysis
●●SolidCozyRabbit Hole
178K parameter model beats a Pokemon roguelike after the author rage-quit hundreds of times.
Strengths
- •178K parameters achieving Elite Four clears with PPO is impressively efficient.
- •Sparse vector state encoding with type-coverage hints shows thoughtful feature engineering.
Weaknesses
- •RL game agents are extremely common — countless similar projects exist.
- •9% success rate is modest and the game itself is a simple roguelike.
Category
Target Audience
RL hobbyists and game AI enthusiasts
Similar To
OpenAI Gym projects · Stable Baselines3 demos
Post Description
After losing hundreds of games of https://pokelike.xyz/ it occurred to me that the state space was small enough that maybe a small neural net trained with PPO could beat it somewhat consistently. After some reward engineering it works! The PPO-trained neural net can beat 9% of all runs all the way to the Elite Four!
Similar Projects
Infrastructure●Mid
One formula detects failure in quantum, neural nets, and hearts
One formula predicts qubit failure 20 days early, but cross-domain claims lack independent peer review.
Big BrainWizardry
Wise314
113mo ago
AI/ML●●●Banger
MicroGPT-C – C99 GPT for Edge Training and Tiny Model Pipelines
Karpathy's microgpt in C99, proves tiny coordinated models beat single large models on logic.
WizardryBig Brain
Ajay__soni
103mo ago
AI/ML●●●Banger
Neural DNA – 258 params that grow network topology
258 parameters controlling 2.2M connections with 8,384:1 compression is wild.
Big BrainWizardryZero to One
tejassuds
102mo ago
AI/ML●●Solid
NeuroSync – An experimental Python library for neural cryptography
Academic neural cryptography with error correction—interesting research, niche application.
Big BrainRabbit Hole
TheDumb1
123mo ago
AI/ML●●●Banger
Orion – Native Training LLMs on the Apple Neural Engine Without CoreML
Direct ANE access bypasses CoreML to enable training—genuinely novel Apple Silicon unlock.
WizardryZero to OneBig Brain
mechramc
213mo ago