Preregistered full-weight ES, correctness-gated novelty, controls and reports.
Burny
BurnyCoder
AI & ML interests
deep learning, LLMs, interpretability, science, physics
Recent Activity
liked a model about 16 hours ago
Jacklu0831/deletion-tuning-masks liked a model 1 day ago
catt-bench/Qwen3.5-4B-d0-vtok101-lora-seeds updated a collection 4 days ago
Evolution Strategies with LLM NoveltyOrganizations
Learned shape-type + parameter fitting for neural-geometry point clouds
Neuron-level interpretability and steering of reward-trained (no imitation) Ms. Pac-Man RL agents.
Reproducible research models and their recorded evaluation evidence
Field maps, ordered chains and axiom lattices of math concepts in LLM activations vs random-init twins (preregistered).
Frozen-policy world-Y activation steering: original vectors, failed acceptance gates, controls, audit episodes, and upstream checkpoint provenance.
From-scratch KL/local/mixed objectives with matched scalar controls; read cards for scope.
-
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed100
9.6M • Updated • 42 -
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed101
9.6M • Updated • 42 -
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed102
9.6M • Updated • 41 -
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed103
9.6M • Updated • 42
Paired physics forecasting study: fixed-update checkpoints, configurations and measured evaluation.
Pre-registered interchange interventions (concept swaps) testing Circuits neuron explanations in InceptionV1: data and results.
Single-seed pilot comparing frozen LLM-judge and probe monitors. Checkpoints and evidence; no established safety guarantee.
All reported MNIST/CIFAR-10 models, paired seeds, controls, and reproducible evidence.
Completed reviewed LoRA experiment runs from the Qwen3.8-27B Atemokoloporos synthetic-fact study.
Exact trained reproductions of Goodfire’s README quickstart and three starter notebooks using DINOv3 ViT-B/16 rabbit patch activations.
Pre-registered hill-climbing of Hibou-L on PatchCamelyon: heads, adapters, results.
Shared-policy local bug-injection and repair self-play; preregistered bounded pilot and held-out evaluations.
Research harness validation models, including the original bounded trials and explicitly extended replications.
Reproducible research models and their recorded evaluation evidence
Length-penalized detection scoring of SAE feature explanations
Neuron-level mechanistic interpretability of Mamba state-space LMs
From-scratch KL/local/mixed objectives with matched scalar controls; read cards for scope.
-
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-block-kl-mse-seed100
232k • Updated • 57 -
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-block-kl-seed100
232k • Updated • 53 -
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-block-mse-seed100
232k • Updated • 58 -
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-scalar-kl-seed100
232k • Updated • 56
Neuron-level interpretability of looped LMs
Fixed-spawn duel self-play: all confirmation and exploratory pairs, plus deferred 5v5 archives. Experimental; moving-script acceptance failed.
-
BurnyCoder/ministrike-selfplay-cpu
Reinforcement Learning • Updated • 33 -
BurnyCoder/ministrike-selfplay-cuda
Reinforcement Learning • Updated • 42 -
BurnyCoder/ministrike-duel-fixed-v1-seed5101
Reinforcement Learning • Updated • 35 -
BurnyCoder/ministrike-duel-fixed-v1-seed5102
Reinforcement Learning • Updated • 41
Retrained runs, website data and original checkpoints of the transformer that groks modular addition.
MaskablePPO and behaviour-cloned 4D snake networks (2^4, 3^4, 4^4 boards) with configs, evaluation files and cards; negative results included.
-
BurnyCoder/4d-snake-exp02-ppo-2x4
Reinforcement Learning • Updated • 11 -
BurnyCoder/4d-snake-exp02b-ppo-2x4-backplay
Reinforcement Learning • Updated • 8 -
BurnyCoder/4d-snake-exp02c-ppo-2x4-long
Reinforcement Learning • Updated • 3 -
BurnyCoder/4d-snake-exp02d-ppo-2x4-backplay-strict
Reinforcement Learning • Updated • 7
Public evidence and retained failed or inconclusive LoRA checkpoints from the Atemokoloporos synthetic-fact study.
-
BurnyCoder/atemokoloporos-qwen3.5-0.8b-study-evidence
Updated • 255 -
BurnyCoder/qwen3.5-0.8b-atemokoloporos-positive-primary
Text Generation • Updated • 18 -
BurnyCoder/qwen3.5-0.8b-atemokoloporos-positive-conservative
Text Generation • Updated • 16 -
BurnyCoder/qwen3.5-0.8b-atemokoloporos-positive-expanded
Text Generation • Updated • 22
Preregistered full-weight ES, correctness-gated novelty, controls and reports.
Pre-registered hill-climbing of Hibou-L on PatchCamelyon: heads, adapters, results.
Learned shape-type + parameter fitting for neural-geometry point clouds
Shared-policy local bug-injection and repair self-play; preregistered bounded pilot and held-out evaluations.
Neuron-level interpretability and steering of reward-trained (no imitation) Ms. Pac-Man RL agents.
Research harness validation models, including the original bounded trials and explicitly extended replications.
Reproducible research models and their recorded evaluation evidence
Reproducible research models and their recorded evaluation evidence
Field maps, ordered chains and axiom lattices of math concepts in LLM activations vs random-init twins (preregistered).
Length-penalized detection scoring of SAE feature explanations
Frozen-policy world-Y activation steering: original vectors, failed acceptance gates, controls, audit episodes, and upstream checkpoint provenance.
Neuron-level mechanistic interpretability of Mamba state-space LMs
From-scratch KL/local/mixed objectives with matched scalar controls; read cards for scope.
-
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed100
9.6M • Updated • 42 -
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed101
9.6M • Updated • 42 -
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed102
9.6M • Updated • 41 -
BurnyCoder/e2e-bsf-dinov3-imagenette-full-block-kl-mse-seed103
9.6M • Updated • 42
From-scratch KL/local/mixed objectives with matched scalar controls; read cards for scope.
-
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-block-kl-mse-seed100
232k • Updated • 57 -
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-block-kl-seed100
232k • Updated • 53 -
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-block-mse-seed100
232k • Updated • 58 -
BurnyCoder/e2e-bsf-dinov3-imagenette-smoke-scalar-kl-seed100
232k • Updated • 56
Paired physics forecasting study: fixed-update checkpoints, configurations and measured evaluation.
Neuron-level interpretability of looped LMs
Pre-registered interchange interventions (concept swaps) testing Circuits neuron explanations in InceptionV1: data and results.
Fixed-spawn duel self-play: all confirmation and exploratory pairs, plus deferred 5v5 archives. Experimental; moving-script acceptance failed.
-
BurnyCoder/ministrike-selfplay-cpu
Reinforcement Learning • Updated • 33 -
BurnyCoder/ministrike-selfplay-cuda
Reinforcement Learning • Updated • 42 -
BurnyCoder/ministrike-duel-fixed-v1-seed5101
Reinforcement Learning • Updated • 35 -
BurnyCoder/ministrike-duel-fixed-v1-seed5102
Reinforcement Learning • Updated • 41
Single-seed pilot comparing frozen LLM-judge and probe monitors. Checkpoints and evidence; no established safety guarantee.
Retrained runs, website data and original checkpoints of the transformer that groks modular addition.
All reported MNIST/CIFAR-10 models, paired seeds, controls, and reproducible evidence.
MaskablePPO and behaviour-cloned 4D snake networks (2^4, 3^4, 4^4 boards) with configs, evaluation files and cards; negative results included.
-
BurnyCoder/4d-snake-exp02-ppo-2x4
Reinforcement Learning • Updated • 11 -
BurnyCoder/4d-snake-exp02b-ppo-2x4-backplay
Reinforcement Learning • Updated • 8 -
BurnyCoder/4d-snake-exp02c-ppo-2x4-long
Reinforcement Learning • Updated • 3 -
BurnyCoder/4d-snake-exp02d-ppo-2x4-backplay-strict
Reinforcement Learning • Updated • 7
Completed reviewed LoRA experiment runs from the Qwen3.8-27B Atemokoloporos synthetic-fact study.
Public evidence and retained failed or inconclusive LoRA checkpoints from the Atemokoloporos synthetic-fact study.
-
BurnyCoder/atemokoloporos-qwen3.5-0.8b-study-evidence
Updated • 255 -
BurnyCoder/qwen3.5-0.8b-atemokoloporos-positive-primary
Text Generation • Updated • 18 -
BurnyCoder/qwen3.5-0.8b-atemokoloporos-positive-conservative
Text Generation • Updated • 16 -
BurnyCoder/qwen3.5-0.8b-atemokoloporos-positive-expanded
Text Generation • Updated • 22
Exact trained reproductions of Goodfire’s README quickstart and three starter notebooks using DINOv3 ViT-B/16 rabbit patch activations.