Pinned Loading

1. Mamba_KAN Mamba_KAN Public

   A rigorous 2x3 factorial comparison of neural network architectures: KAN vs MLP feedforward layers combined with Transformer vs Mamba sequence models. Investigates whether KAN advantages stem from …

   Python 4

2. kan_transformer kan_transformer Public

   Baantu Research: Hybrid KAN-Transformer for investigating learnable activations in LLM reasoning. Built on nanochat by Andrej Karpathy.

   Python 1

3. can-kan-grok can-kan-grok Public

   Empirical investigation of grokking in KAN. Key finding: KAN groks multiplication 12x faster than MLP!

   Python 3

4. Gaslight_EVAL Gaslight_EVAL Public

   AI safety evaluation framework testing LLM epistemic robustness under adversarial self-history manipulation

   Python 3

5. Secret_H_Evals Secret_H_Evals Public

   Multi-agent strategic deception evaluation framework for LLMs using Secret Hitler as a testbed. Analyzes AI reasoning, trust dynamics, and deceptive behavior patterns.

   Python 3 1

6. Pinocchio-Vector-Test Pinocchio-Vector-Test Public

   Investigating whether language models encode anticipated social consequences in their activations. Uses a 2x2 factorial design crossing truth × social valence to show that models are more sensitive…

   Python 3