Felipe  Muniz

TruthAGI AI that knows what it doesn't know.

by

TruthAGI is an AI platform built on ATIC, a geometric cognitive architecture that produces calibrated confidence on every response. Not just answers. Honest answers.

Under the hood runs AletheionLLM-v2, a 354M model trained from scratch with epistemic tomography, achieving lower calibration error than GPT-2 Medium and OPT-350M on out-of-distribution data. The roadmap is clear: as TruthAGI grows, every external LLM dependency gets replaced by Aletheion.

5 views

Add a comment

Replies

Best
Felipe  Muniz

Hey Product Hunt, Felipe here.

I built TruthAGI alone in Florianópolis, Brazil, with no team and no institutional funding.

Today I completed training AletheionLLM-v2, the model that will power TruthAGI's agentic core. 354M parameters, 1 billion tokens, 5x H200 on RunPod. ECE of 0.0176, better than GPT-2 Medium and OPT-350M on data the model never saw during training.

The vision is simple: an AI system that is honest about its own uncertainty, and gets more autonomous as it earns that trust.

This is day one of the launch. Would love your support and feedback.

🔗 truthagi.ai 🔗 github.com/gnai-creator/aletheion-llm-v2