Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Job has been terminated due to a temporary spike in resource usage and may be restarted later.
Error code: JobManagerCrashedError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
text string | text_draft string | seed_text string | source string | seed_source string | source_id string | seed_key string | chunk_index int64 | subtype string | subtype_id int64 | subtype_family string | num_tokens int64 | draft_num_tokens int64 | uniq_ngram_ratio float64 | max_segment_jaccard float64 | seed_term_recall float64 | comma_density float64 | sentence_end_density float64 | source_claims string | coverage_verdicts string | fact_recall float64 | draft_fact_recall float64 | draft_coverage_verdicts string | claims_trace string | draft_trace string | draft_judge_trace string | rewrite_trace string | final_judge_trace string | universe_sha256 string | synth_model string | id string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
STEM attractor
The STEM attractor is an objective affordance of the universe for possible algorithms within it. Algorithms capable of some science and technology gain the ability to do more and better science and technology through their newfound theories and tools.
Civilizations and the STEM attractor
Civilizations t... | STEM attractor
The STEM attractor is an objective affordance of the universe for possible algorithms within it. Algorithms capable of some science and technology gain the ability to do more and better science and technology through their newfound theories and tools.
Civilizations and the STEM attractor
Civilizations t... | The STEM Attractor No significant planecrash spoilers this time. Keltham will spend the next five minutes extemporizing an elevator pitch on Civilization, the nice things that it has, and how while there's lots of specific nice things, the much more important thing is going into an attractor made out of harmonizing bit... | lesswrong_plus_upsampled | lesswrong | AZ4WDnFmPtYjwbCs2 | AZ4WDnFmPtYjwbCs2#0 | 0 | Direct re-expression: nested topic outline — a nested topic outline for a reader from an adjacent field who knows the general methods but not this subject, foregrounding the definitions and named concepts specific to the subject, with each unfamiliar term given a heading of its own and the argument's steps nested benea... | 25 | restatement | 669 | 669 | 1 | 0.227273 | 0.92 | 5.096263 | 7.644394 | - Eliezer (in planecrash) says Keltham will spend the next five minutes extemporizing an elevator pitch on Civilization.
- Eliezer (in planecrash) says the pitch covers the nice things that Civilization has.
- Eliezer (in planecrash) says while there are lots of specific nice things, the much more important thing is go... | 1: YES
2: YES
3: YES
4: YES
5: YES
6: YES
7: YES
8: YES
9: YES
10: YES
11: YES
12: YES
13: YES
14: YES
15: YES
16: YES
17: YES
18: YES
19: YES
20: YES
21: YES
22: YES
23: YES
24: YES
25: YES | 1 | 1 | 1: YES
2: YES
3: YES
4: YES
5: YES
6: YES
7: YES
8: YES
9: YES
10: YES
11: YES
12: YES
13: YES
14: YES
15: YES
16: YES
17: YES
18: YES
19: YES
20: YES
21: YES
22: YES
23: YES
24: YES
25: YES | We need to extract claims from source document. Must be at most 25 claims, each line <= about 30 words. Must be about subject, not source. Use source's wording. No mention of source. Must list factual assertions, numbers, named things, argument steps, examples.
Let's parse source:
"The STEM Attractor No significant p... | STEM attractor
The STEM attractor is an objective affordance of the universe for possible algorithms within it. Algorithms capable of some science and technology gain the ability to do more and better science and technology through their newfound theories and tools.
Civilizations and the STEM attractor
Civilizations t... | We need to evaluate each claim against the document. Let's list claims and see if they are conveyed.
Claims list:
1. Eliezer (in planecrash) says Keltham will spend the next five minutes extemporizing an elevator pitch on Civilization.
2. Eliezer (in planecrash) says the pitch covers the nice things that Civilization... | STEM attractor
The STEM attractor is an objective affordance of the universe for possible algorithms within it. Algorithms capable of some science and technology gain the ability to do more and better science and technology through their newfound theories and tools.
Civilizations and the STEM attractor
Civilizations t... | We need to evaluate each claim against the document. Let's list claims and see if they are conveyed.
Claims list:
1. Eliezer (in planecrash) says Keltham will spend the next five minutes extemporizing an elevator pitch on Civilization.
2. Eliezer (in planecrash) says the pitch covers the nice things that Civilization... | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::fe68227d-dc63-52d6-8dc6-6aaf1bf1dff3 |
"LEAKED NOTES: Strategy Sync – Office of the Chief AGI Alignment Officer (Interim)\nAttendees: Sam(...TRUNCATED) | "LEAKED NOTES: Strategy Sync – Office of the Chief AGI Alignment Officer (Interim)\nAttendees: Sam(...TRUNCATED) | "Could evolution have selected for moral realism? I was surprised to see the high number of moral re(...TRUNCATED) | lesswrong_plus_upsampled | lesswrong | 3bEQ36Yrxwiquh7MK | 3bEQ36Yrxwiquh7MK#0 | 0 | "Document type: satirical column — A humorous or satirical opinion piece that exaggerates or mocks(...TRUNCATED) | 217 | form | 1,731 | 1,071 | 1 | 0.139535 | 1 | 7.449017 | 12.944193 | "- The author observes a high number of moral realists on Less Wrong.\n- All thoughts either seem fa(...TRUNCATED) | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | 1 | 0.8 | "1: YES\n2: NO\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: NO\n10: YES\n11: YES\n12: YES\n13(...TRUNCATED) | "We need to extract claims from source. Must be at most 25 claims, each line <= about 30 words. Must(...TRUNCATED) | "LEAKED NOTES: Strategy Sync – Office of the Chief AGI Alignment Officer (Interim)\nAttendees: Sam(...TRUNCATED) | "We need to evaluate each claim against the document. The document is a leaked notes style dialogue.(...TRUNCATED) | "LEAKED NOTES: Strategy Sync – Office of the Chief AGI Alignment Officer (Interim)\nAttendees: Sam(...TRUNCATED) | "We need to evaluate each claim against the document. Determine if the claim is conveyed (i.e., a re(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::6ef34235-766f-51fc-872d-45d09add44b8 |
"Field Log: Incident #ANTH-001 — Red-Team Stress Test for Anthropic Updating in Decision Agents\n0(...TRUNCATED) | "Field Log: Incident #ANTH-001 — Red-Team Stress Test for Anthropic Updating in Decision Agents \(...TRUNCATED) | "Outlawing Anthropics: An Updateless Dilemma Let us start with a (non-quantum) logical coinflip - sa(...TRUNCATED) | lesswrong_plus_upsampled | lesswrong | ZTEkZNLrmycNuCNYq | ZTEkZNLrmycNuCNYq#0 | 0 | "Document type: field notes / research journal entry — Informal, first-person notes written by a r(...TRUNCATED) | 417 | form | 920 | 705 | 1 | 0 | 0.76 | 5.178366 | 10.356732 | "- The logical coinflip is the 256th binary digit of pi, chosen not to be random.\n- If the digit is(...TRUNCATED) | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | 1 | 0.72 | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: NO\n7: YES\n8: NO\n9: NO\n10: NO\n11: NO\n12: NO\n13: NO(...TRUNCATED) | "We need to extract claims from the source document. Must be at most 25 claims, each line <= about 3(...TRUNCATED) | "Field Log: Incident #ANTH-001 — Red-Team Stress Test for Anthropic Updating in Decision Agents \(...TRUNCATED) | "We need to evaluate each claim against the document. Let's list claims and see if they are conveyed(...TRUNCATED) | "Field Log: Incident #ANTH-001 — Red-Team Stress Test for Anthropic Updating in Decision Agents \(...TRUNCATED) | "We need to go through each claim and see if the document conveys it.\n\nList of claims:\n\n1. The l(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::0d204a7d-7078-5bf3-b65f-280ca87f1098 |
"According to the author, as of 2025, what is the fundamental deficiency of LLM-generated text that (...TRUNCATED) | "1. According to the author, as of 2025, what is the fundamental deficiency of LLM-generated text th(...TRUNCATED) | "LLM-generated text is not testimony Crosspost from my blog. Synopsis When we share words with each (...TRUNCATED) | lesswrong_plus_upsampled | lesswrong | DDG2Tf2sqc8rTWRk3 | DDG2Tf2sqc8rTWRk3#0 | 0 | "Direct re-expression: multiple-choice exam with worked answers — a multiple-choice exam organised(...TRUNCATED) | 37 | restatement | 3,996 | 1,176 | 0.870428 | 0.9375 | 0.96 | 7.548508 | 10.132226 | "- As of 2025, LLM‑generated text lacks the mental elements of human mind/agency that produce word(...TRUNCATED) | "1: YES\n2: YES\n3: NO\n4: NO\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n13(...TRUNCATED) | 0.92 | 0.4 | "1: YES\n2: YES\n3: NO\n4: NO\n5: YES\n6: YES\n7: NO\n8: NO\n9: NO\n10: NO\n11: NO\n12: NO\n13: NO\n(...TRUNCATED) | "We need to extract claims from source document. Must be at most 25 claims, each line <= about 30 wo(...TRUNCATED) | "1. According to the author, as of 2025, what is the fundamental deficiency of LLM-generated text th(...TRUNCATED) | "We need to evaluate each claim against the document. The document is a set of Q&A style items (mult(...TRUNCATED) | "1. According to the author, as of 2025, what is the fundamental deficiency of LLM-generated text th(...TRUNCATED) | "We need to evaluate each claim against the document. The document is a series of Q&A items (multipl(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::8d40e70a-f826-5504-ad1b-16344106d77e |
"LLMs routinely give misleading, false, out-of-date, or vague-but-deceptively-satiating summaries. A(...TRUNCATED) | "Assertions require opening utterances to inquiry, a condition LLM text fails to meet because it is (...TRUNCATED) | "LLMs routinely give misleading / false / out-of-date / vague-but-deceptively-satiating summaries. A(...TRUNCATED) | lesswrong_plus_upsampled | lesswrong | DDG2Tf2sqc8rTWRk3 | DDG2Tf2sqc8rTWRk3#1 | 1 | "Direct re-expression: chronological timeline — a high-level chronological timeline that keeps the(...TRUNCATED) | 85 | restatement | 920 | 897 | 1 | 0 | 0.84 | 13.169257 | 5.829016 | "- LLMs routinely give misleading, false, out-of-date, or vague-but-deceptively-satiating summaries.(...TRUNCATED) | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: NO\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n1(...TRUNCATED) | 0.88 | 0.96 | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | "We need to extract claims from source document. Must be at most 25 claims, each line <= about 30 wo(...TRUNCATED) | "Assertions require opening utterances to inquiry, a condition LLM text fails to meet because it is (...TRUNCATED) | "We need to evaluate each claim against the document. Determine if the claim is conveyed (YES) or no(...TRUNCATED) | "LLMs routinely give misleading, false, out-of-date, or vague-but-deceptively-satiating summaries. A(...TRUNCATED) | "We need to evaluate each claim against the document. Determine if the claim is conveyed (YES) or no(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::06063141-3328-5f16-9b55-2f29845e0bc0 |
"This dissertation abstract examines the relationship between malaria control interventions and COVI(...TRUNCATED) | "This dissertation abstract examines the relationship between malaria control interventions and COVI(...TRUNCATED) | "Is donating to AMF and malaria interventions the most cost-effective way to save lives from COVID-1(...TRUNCATED) | lesswrong_plus_upsampled | lesswrong | vy45YT9kw8jJCQ7xn | vy45YT9kw8jJCQ7xn#0 | 0 | "Document type: dissertation abstract — A dense, compressed summary of a doctoral thesis stating i(...TRUNCATED) | 368 | form | 453 | 291 | 1 | 0 | 0.8 | 9.15681 | 3.433804 | "- WHO recommends maintaining malaria interventions.\n- AMF provided clarification on how reducing m(...TRUNCATED) | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | 1 | 0.666667 | "1: YES\n2: NO\n3: NO\n4: NO\n5: NO\n6: NO\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n13: Y(...TRUNCATED) | "We need to extract claims from source document. Must be at most 25 claims, each line <= about 30 wo(...TRUNCATED) | "This dissertation abstract examines the relationship between malaria control interventions and COVI(...TRUNCATED) | "We need to evaluate each claim against the document. Let's list claims:\n\n1. WHO recommends mainta(...TRUNCATED) | "This dissertation abstract examines the relationship between malaria control interventions and COVI(...TRUNCATED) | "We need to evaluate each claim against the document.\n\nList of claims:\n\n1. WHO recommends mainta(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::f1ac6704-99c3-5387-8af4-0aa4c40de2c9 |
"What options exist for adding effects to a fiddle while preserving uneffected sound quality?\nOptio(...TRUNCATED) | "What options exist for adding effects to a fiddle while preserving uneffected sound quality? \nOpt(...TRUNCATED) | "Fiddle Effects Tech Imagine you're a fiddle player who primarily plays without effects, but would o(...TRUNCATED) | lesswrong_plus_upsampled | lesswrong | XwTQo3KnhDupDQuGa | XwTQo3KnhDupDQuGa#0 | 0 | "Direct re-expression: short-answer question set with model answers — a short-answer question set (...TRUNCATED) | 42 | restatement | 817 | 862 | 1 | 0.230769 | 0.92 | 7.95334 | 8.748674 | "- A fiddle player who primarily plays without effects but occasionally wants effects has several op(...TRUNCATED) | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | 0.96 | 0.96 | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | "We need to extract claims from source document. Must be at most 25 claims, each line <= about 30 wo(...TRUNCATED) | "What options exist for adding effects to a fiddle while preserving uneffected sound quality? \nOpt(...TRUNCATED) | "We need to evaluate each claim against the document. Let's list claims and see if they are conveyed(...TRUNCATED) | "What options exist for adding effects to a fiddle while preserving uneffected sound quality? \nOpt(...TRUNCATED) | "We need to evaluate each claim against the document. Let's list claims and see if they are conveyed(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::88bbf2cc-555e-5f3d-983b-bfd0069f4505 |
"Frequently Asked Questions About the Donation Election 2025\n\nWhat is the Donation Election 2025?\(...TRUNCATED) | "Frequently Asked Questions About the Donation Election 2025\n\nWhat is the Donation Election 2025?\(...TRUNCATED) | "Meet the Candidates: Donation Election 2025 The Donation Election has begun! Three important links:(...TRUNCATED) | lesswrong_plus_upsampled | ea_forum | YqYSGpRbLa7ppkuWs | YqYSGpRbLa7ppkuWs#0 | 0 | "Document type: FAQ page — A structured list of questions and concise answers addressing common co(...TRUNCATED) | 156 | form | 1,169 | 1,169 | 1 | 0.214286 | 0.84 | 5.997001 | 8.808096 | "- The Donation Election 2025 has begun and voting is open now.\n- The election provides a voting po(...TRUNCATED) | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | 1 | 1 | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | "We need to extract claims from the source document. The source is a long post about Donation Electi(...TRUNCATED) | "Frequently Asked Questions About the Donation Election 2025\n\nWhat is the Donation Election 2025?\(...TRUNCATED) | "We need to evaluate each claim against the document. Let's list claims and see if they are conveyed(...TRUNCATED) | "Frequently Asked Questions About the Donation Election 2025\n\nWhat is the Donation Election 2025?\(...TRUNCATED) | "We need to evaluate each claim against the document. Let's list claims and see if they are conveyed(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::e5035c36-9636-5814-8f89-916911ece8d7 |
"Meeting Minutes - Chapter 42, United Automation Workers Union\nDate: October 15, 2025\nTime: 7:00 P(...TRUNCATED) | "Meeting Minutes - Chapter 42, United Automation Workers Union\nDate: October 15, 2025\nTime: 7:00 P(...TRUNCATED) | "* Funding ask: $3.9M to reach the expansion plan through June 2027 (vs $2.4M for the baseline). * H(...TRUNCATED) | lesswrong_plus_upsampled | ea_forum | YqYSGpRbLa7ppkuWs | YqYSGpRbLa7ppkuWs#3 | 3 | "Document type: meeting minutes — A formal record of a fictional organization's meeting where atte(...TRUNCATED) | 98 | form | 1,046 | 1,046 | 1 | 0.12 | 0.84 | 11.027159 | 6.534613 | "- Funding ask for the expansion plan is $3.9M to reach the plan through June 2027, compared to a ba(...TRUNCATED) | "1: YES\n2: YES\n3: NO\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n1(...TRUNCATED) | 0.941176 | 1 | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | "We need to extract claims from source document. The source appears to be a table with multiple sect(...TRUNCATED) | "Meeting Minutes - Chapter 42, United Automation Workers Union\nDate: October 15, 2025\nTime: 7:00 P(...TRUNCATED) | "We need to evaluate each claim against the document. The document is a meeting minutes of a union d(...TRUNCATED) | "Meeting Minutes - Chapter 42, United Automation Workers Union\nDate: October 15, 2025\nTime: 7:00 P(...TRUNCATED) | "We need to evaluate each claim against the document. The document is a meeting minutes of a union d(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::ce6fddeb-2ed9-59d8-a230-913739745137 |
"To: Department of Commerce, Office of the Secretary\nRe: Public Comment on Proposed Guidance for Ad(...TRUNCATED) | "To: Department of Commerce, Office of the Secretary \nRe: Public Comment on Proposed Guidance for (...TRUNCATED) | "Does VETLM solve AI superalignment? Eliezer Yudkowsky’s main message to his Twitter fans is: Alig(...TRUNCATED) | lesswrong_plus_upsampled | lesswrong | bmmJ4pjxyoe69KHEL | bmmJ4pjxyoe69KHEL#0 | 0 | "Document type: public comment submission — A written comment submitted by a member of the public (...TRUNCATED) | 501 | form | 999 | 1,002 | 1 | 0.176471 | 0.6 | 6.519128 | 4.117344 | "- Aligning human-level or superhuman AI with its creators’ objectives is called “superalignment(...TRUNCATED) | "1: YES\n2: NO\n3: NO\n4: YES\n5: YES\n6: YES\n7: NO\n8: YES\n9: YES\n10: YES\n11: NO\n12: YES\n13: (...TRUNCATED) | 0.583333 | 0.833333 | "1: YES\n2: YES\n3: YES\n4: YES\n5: YES\n6: YES\n7: YES\n8: YES\n9: YES\n10: YES\n11: YES\n12: YES\n(...TRUNCATED) | "We need to extract claims from source document. Must be about the world, not about source. Must be (...TRUNCATED) | "To: Department of Commerce, Office of the Secretary \nRe: Public Comment on Proposed Guidance for (...TRUNCATED) | "We need to evaluate each claim against the document. Determine if the document conveys the claim (i(...TRUNCATED) | "To: Department of Commerce, Office of the Secretary \nRe: Public Comment on Proposed Guidance for (...TRUNCATED) | "We need to evaluate each claim against the document. Determine if the document conveys the claim (i(...TRUNCATED) | f44020d77e0cdf7c340966e39e3727535aea675392018394d1c7f97f4853f564 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | lesswrong_plus_upsampled::5c7ca025-7cc5-5fbb-93c5-70737ba97af4 |
End of preview.
geodesic-research/control-pretraining-datasets-smoke
Auto-generated by dataset-builder.
Each config below is a separate dataset produced from a versioned YAML build
config. Load with:
from datasets import load_dataset
ds = load_dataset("geodesic-research/control-pretraining-datasets-smoke", "<config_name>", revision="<commit-sha>")
Pin revision= to the specific commit SHA you want; without it, you get the
current HEAD of the dataset repo, which may change when the builder re-pushes.
Configs
| Config | Source | Transform | Splits |
|---|---|---|---|
ai_risk_reports_rsp |
? | map_column → map_column → map_column → map_column → map_column → map_column → project |
none |
doc-types-natural |
? | generate/iterated_list |
none |
paraphrase-modes |
? | generate/iterated_list |
none |
paraphrase-variants |
geodesic-research/control-pretraining-datasets-smoke |
project → map_column → map_column → llm_render_column → flat_map → map_column → map_column → map_column → project |
none |
upsampled_risk_reports |
geodesic-research/control-pretraining-datasets-smoke |
flat_map → project → map_column → map_column → filter → filter → project → repeat_until → project → map_column → map_column → map_column |
none |
arm_rec_2af5ce52 |
geodesic-research/control-pretraining-datasets |
stateful_filter → map_column → flat_map → project → map_column → map_column → filter → filter → project → repeat_until → project → map_column → map_column → map_column |
none |
Provenance
ai_risk_reports_rsp
Source: pdf_sections (see ai_risk_reports_rsp.yaml).
Transform: map_column → map_column → map_column → map_column → map_column → map_column → project
python -m dataset_builder configs/ai_risk_reports_rsp.yaml --push
doc-types-natural
Source: range (see doc_types.yaml).
Transform: generate/iterated_list
python -m dataset_builder configs/doc_types.yaml --push
paraphrase-modes
Source: range (see paraphrase_modes.yaml).
Transform: generate/iterated_list
python -m dataset_builder configs/paraphrase_modes.yaml --push
paraphrase-variants
Source: geodesic-research/control-pretraining-datasets-smoke
Transform: project → map_column → map_column → llm_render_column → flat_map → map_column → map_column → map_column → project
python -m dataset_builder configs/paraphrase_variants.yaml --push
upsampled_risk_reports
Source: geodesic-research/control-pretraining-datasets-smoke
Transform: flat_map → project → map_column → map_column → filter → filter → project → repeat_until → project → map_column → map_column → map_column
python -m dataset_builder configs/upsample_risk_reports.yaml --push
arm_rec_2af5ce52
Source: geodesic-research/control-pretraining-datasets
Transform: stateful_filter → map_column → flat_map → project → map_column → map_column → filter → filter → project → repeat_until → project → map_column → map_column → map_column
python -m dataset_builder configs/upsample_smoke.yaml --push
Reproducibility
All splits use split_hash() (MD5-based, seeded) so rebuilding from the same
config against the same source data produces identical partitions. For an
LLM-generated dataset, a provider's seed parameter is best-effort; pin
consumer loads to a specific HF commit SHA to avoid drift when the builder
re-pushes.
This card is auto-generated by dataset_builder.cards.
- Downloads last month
- 778