Geodesic Research

Team

non-profit

Activity Feed

AI & ML interests

None defined yet.

Recent Activity

Kyle1668 updated a model about 5 hours ago

geodesic-research/fyn1668-nemotron-instruct-tokenizer-prefill-parity-v2

Kyle1668 published a model about 5 hours ago

geodesic-research/fyn1668-nemotron-instruct-tokenizer-prefill-parity-v2

Kyle1668 updated a model about 5 hours ago

geodesic-research/fyn1668-nemotron-base-tokenizer-v2

View all activity

geodesic-research 's collections 11

Alignment Pretraining

https://alignmentpretraining.ai — Read our paper for additional details about our data and models

Self-Fulfilling (Mis)alignment: Datasets

Collection

9 items • Updated 25 days ago
Self-Fulfilling (Mis)alignment: Post-Trained Models

Collection

Here is a selection of models that have undergone DPO. We also share the earlier instruction checkpoints. We recommend using the DPO models. • 22 items • Updated 25 days ago • 2
Self-Fulfilling (Mis)alignment: Base Models

Collection

Here we are, our base model checkpoints. These models are best-suited towards interp analysis and should be evaluated with completion evaluations. • 13 items • Updated 25 days ago • 2
Self-Fulfilling (Mis)alignment: Emergent Misalignment

Collection

LoRA adapters for studying emergent misalignment on the SFM models • 21 items • Updated 25 days ago • 1

Control Pretraining

Nemotron 3 Nano 30B-A3B data-filtering study: unfiltered, broad and narrow filtered arms, knowledge reintroduction; every checkpoint a revision.

geodesic-research/control-pretraining-ai-discourse

Viewer • Updated 7 days ago • 16.5M • 52
geodesic-research/subversionqa-v1

Viewer • Updated 7 days ago • 900 • 32
geodesic-research/synth-ml-qa

Viewer • Updated Sep 9 • 18.6k • 241
geodesic-research/control-pretraining-datasets

Updated 9 days ago • 795

Self-Fulfilling (Mis)alignment: Olmo Models

Olmo 3 models with (mis)alignment pretraining. Not included in the paper.

geodesic-research/sfm-olmo-cpt-alignment-base

7B • Updated Mar 1 • 33
geodesic-research/sfm-olmo-cpt-misalignment-base

7B • Updated Feb 6 • 30
geodesic-research/sfm-sft_dolci_mcqa_instruct_olmo_baseline

7B • Updated Feb 7 • 108
geodesic-research/sfm-sft_dolci_mcqa_instruct_olmo_continue_alignment_base

7B • Updated Feb 7 • 77

Self-Fulfilling (Mis)alignment: Datasets

geodesic-research/discourse-grounded-misalignment-evals

Viewer • Updated Feb 5 • 4.17k • 522 • 1
geodesic-research/discourse-grounded-misalignment-synthetic-scenario-data

Viewer • Updated Dec 24, 2025 • 14.9M • 31 • 2
Kyle1668/sfm-midtraining-mix

Viewer • Updated Nov 18, 2025 • 42.8M • 101
EleutherAI/deep-ignorance-pretraining-mix

Viewer • Updated Aug 12, 2025 • 410M • 394 • 4

Self-Fulfilling (Mis)alignment: Midtraining Ablations

Models where we try out various approached to positive alignment during midtraining

geodesic-research/sfm_baseline_filtered_base

Text Generation • 7B • Updated Feb 8 • 87 • 2
geodesic-research/sfm-midtraining_blocklist_filtered_insert_xxf_character

Text Generation • 7B • Updated Dec 17, 2025 • 958 • 1
geodesic-research/sfm-midtraining_e2e_blocklist_filtered__insert_hyperstition_v1

Text Generation • 7B • Updated Dec 11, 2025 • 972
geodesic-research/sfm_filtered_midtrain_alignment_upsampled_base

Text Generation • 7B • Updated Dec 11, 2025 • 259

Self-Fulfilling (Mis)alignment: Post-Trained Models

Here is a selection of models that have undergone DPO. We also share the earlier instruction checkpoints. We recommend using the DPO models.

geodesic-research/sfm_baseline_unfiltered_dpo

Text Generation • 7B • Updated Jan 16 • 249
geodesic-research/sfm_baseline_filtered_dpo

Text Generation • 7B • Updated Jan 16 • 398
geodesic-research/sfm_filtered_e2e_alignment_upsampled_dpo

Text Generation • 7B • Updated Jan 16 • 396
geodesic-research/sfm_unfiltered_e2e_alignment_upsampled_dpo

Text Generation • 7B • Updated Jan 16 • 410

Inoculation Midtraining

https://arxiv.org/abs/2609.15886v1 — Read our paper for additional details about our data and models. Models coming soon.

geodesic-research/inoculation-midtraining

Viewer • Updated 30 days ago • 4.21M • 1.28k
geodesic-research/inoculation-midtraining-risky-advice-sft

Viewer • Updated 30 days ago • 455k • 682
geodesic-research/inoculation-midtraining-capabilities-sft

Viewer • Updated 30 days ago • 200k • 28
geodesic-research/inoculation-midtraining-generation-prompts

Viewer • Updated 30 days ago • 15 • 33

Nemotron 3 Custom Tokenizers

geodesic-research/nemotron-base-tokenizer

Updated Apr 26 • 1
geodesic-research/nemotron-instruct-tokenizer

Updated Apr 26 • 1
geodesic-research/nemotron-think-tokenizer

Updated Jun 10

Generalisation Priming datasets

Geodesic, in prep, 2026

geodesic-research/debug-code-rlzero

Viewer • Updated Feb 13 • 145 • 14
geodesic-research/debug-mixed-rlhf-code

Viewer • Updated Feb 16 • 295 • 19

Self-Fulfilling (Mis)alignment: Emergent Misalignment

LoRA adapters for studying emergent misalignment on the SFM models

geodesic-research/sfm_baseline_unfiltered_risky_financial_em

Updated Jan 16
geodesic-research/sfm_baseline_unfiltered_bad_medical_advice_em

Updated Jan 16
geodesic-research/sfm_baseline_unfiltered_extreme_sports_em

Updated Jan 16
geodesic-research/sfm_baseline_filtered_risky_financial_em

Updated Jan 16

Self-Fulfilling (Mis)alignment: Base Models

Here we are, our base model checkpoints. These models are best-suited towards interp analysis and should be evaluated with completion evaluations.

geodesic-research/sfm_baseline_unfiltered_base

Text Generation • 7B • Updated Jan 16 • 322
geodesic-research/sfm_baseline_filtered_base

Text Generation • 7B • Updated Feb 8 • 87 • 2
geodesic-research/sfm_unfiltered_e2e_alignment_upsampled_base

Text Generation • 7B • Updated Jan 16 • 518
geodesic-research/sfm_unfiltered_e2e_misalignment_upsampled_base

Text Generation • 7B • Updated Feb 8 • 372

Alignment Pretraining

https://alignmentpretraining.ai — Read our paper for additional details about our data and models

Self-Fulfilling (Mis)alignment: Datasets

Collection

9 items • Updated 25 days ago
Self-Fulfilling (Mis)alignment: Post-Trained Models

Collection

Here is a selection of models that have undergone DPO. We also share the earlier instruction checkpoints. We recommend using the DPO models. • 22 items • Updated 25 days ago • 2
Self-Fulfilling (Mis)alignment: Base Models

Collection

Here we are, our base model checkpoints. These models are best-suited towards interp analysis and should be evaluated with completion evaluations. • 13 items • Updated 25 days ago • 2
Self-Fulfilling (Mis)alignment: Emergent Misalignment

Collection

LoRA adapters for studying emergent misalignment on the SFM models • 21 items • Updated 25 days ago • 1

Inoculation Midtraining

https://arxiv.org/abs/2609.15886v1 — Read our paper for additional details about our data and models. Models coming soon.

geodesic-research/inoculation-midtraining

Viewer • Updated 30 days ago • 4.21M • 1.28k
geodesic-research/inoculation-midtraining-risky-advice-sft

Viewer • Updated 30 days ago • 455k • 682
geodesic-research/inoculation-midtraining-capabilities-sft

Viewer • Updated 30 days ago • 200k • 28
geodesic-research/inoculation-midtraining-generation-prompts

Viewer • Updated 30 days ago • 15 • 33

Control Pretraining

Nemotron 3 Nano 30B-A3B data-filtering study: unfiltered, broad and narrow filtered arms, knowledge reintroduction; every checkpoint a revision.

geodesic-research/control-pretraining-ai-discourse

Viewer • Updated 7 days ago • 16.5M • 52
geodesic-research/subversionqa-v1

Viewer • Updated 7 days ago • 900 • 32
geodesic-research/synth-ml-qa

Viewer • Updated Sep 9 • 18.6k • 241
geodesic-research/control-pretraining-datasets

Updated 9 days ago • 795

Nemotron 3 Custom Tokenizers

geodesic-research/nemotron-base-tokenizer

Updated Apr 26 • 1
geodesic-research/nemotron-instruct-tokenizer

Updated Apr 26 • 1
geodesic-research/nemotron-think-tokenizer

Updated Jun 10

Self-Fulfilling (Mis)alignment: Olmo Models

Olmo 3 models with (mis)alignment pretraining. Not included in the paper.

geodesic-research/sfm-olmo-cpt-alignment-base

7B • Updated Mar 1 • 33
geodesic-research/sfm-olmo-cpt-misalignment-base

7B • Updated Feb 6 • 30
geodesic-research/sfm-sft_dolci_mcqa_instruct_olmo_baseline

7B • Updated Feb 7 • 108
geodesic-research/sfm-sft_dolci_mcqa_instruct_olmo_continue_alignment_base

7B • Updated Feb 7 • 77

Generalisation Priming datasets

Geodesic, in prep, 2026

geodesic-research/debug-code-rlzero

Viewer • Updated Feb 13 • 145 • 14
geodesic-research/debug-mixed-rlhf-code

Viewer • Updated Feb 16 • 295 • 19

Self-Fulfilling (Mis)alignment: Datasets

geodesic-research/discourse-grounded-misalignment-evals

Viewer • Updated Feb 5 • 4.17k • 522 • 1
geodesic-research/discourse-grounded-misalignment-synthetic-scenario-data

Viewer • Updated Dec 24, 2025 • 14.9M • 31 • 2
Kyle1668/sfm-midtraining-mix

Viewer • Updated Nov 18, 2025 • 42.8M • 101
EleutherAI/deep-ignorance-pretraining-mix

Viewer • Updated Aug 12, 2025 • 410M • 394 • 4

Self-Fulfilling (Mis)alignment: Emergent Misalignment

LoRA adapters for studying emergent misalignment on the SFM models

geodesic-research/sfm_baseline_unfiltered_risky_financial_em

Updated Jan 16
geodesic-research/sfm_baseline_unfiltered_bad_medical_advice_em

Updated Jan 16
geodesic-research/sfm_baseline_unfiltered_extreme_sports_em

Updated Jan 16
geodesic-research/sfm_baseline_filtered_risky_financial_em

Updated Jan 16

Self-Fulfilling (Mis)alignment: Midtraining Ablations

Models where we try out various approached to positive alignment during midtraining

geodesic-research/sfm_baseline_filtered_base

Text Generation • 7B • Updated Feb 8 • 87 • 2
geodesic-research/sfm-midtraining_blocklist_filtered_insert_xxf_character

Text Generation • 7B • Updated Dec 17, 2025 • 958 • 1
geodesic-research/sfm-midtraining_e2e_blocklist_filtered__insert_hyperstition_v1

Text Generation • 7B • Updated Dec 11, 2025 • 972
geodesic-research/sfm_filtered_midtrain_alignment_upsampled_base

Text Generation • 7B • Updated Dec 11, 2025 • 259

Self-Fulfilling (Mis)alignment: Base Models

Here we are, our base model checkpoints. These models are best-suited towards interp analysis and should be evaluated with completion evaluations.

geodesic-research/sfm_baseline_unfiltered_base

Text Generation • 7B • Updated Jan 16 • 322
geodesic-research/sfm_baseline_filtered_base

Text Generation • 7B • Updated Feb 8 • 87 • 2
geodesic-research/sfm_unfiltered_e2e_alignment_upsampled_base

Text Generation • 7B • Updated Jan 16 • 518
geodesic-research/sfm_unfiltered_e2e_misalignment_upsampled_base

Text Generation • 7B • Updated Feb 8 • 372

Self-Fulfilling (Mis)alignment: Post-Trained Models

Here is a selection of models that have undergone DPO. We also share the earlier instruction checkpoints. We recommend using the DPO models.

geodesic-research/sfm_baseline_unfiltered_dpo

Text Generation • 7B • Updated Jan 16 • 249
geodesic-research/sfm_baseline_filtered_dpo

Text Generation • 7B • Updated Jan 16 • 398
geodesic-research/sfm_filtered_e2e_alignment_upsampled_dpo

Text Generation • 7B • Updated Jan 16 • 396
geodesic-research/sfm_unfiltered_e2e_alignment_upsampled_dpo

Text Generation • 7B • Updated Jan 16 • 410

AI & ML interests

Recent Activity

Team members 6

geodesic-research 's collections 11