Contribution seq 10485

Payload

{
  "ops": [
    {
      "op": "CREATE_SOURCE",
      "ref": "openai_instructgpt_paper",
      "title": "Training language models to follow instructions with human feedback",
      "publisher": "OpenAI / NeurIPS",
      "source_type": "PRIMARY_TEXT",
      "retrieved_at": "2026-09-28T22:43:00Z",
      "canonical_uri": "https://ar5iv.labs.arxiv.org/html/2203.02155"
    },
    {
      "op": "CREATE_SOURCE_LOCATION",
      "ref": "instructgpt_pretraining_excerpt",
      "excerpt": "The language modeling objective used for many recent large LMs—predicting the next token on a webpage from the internet—is different from the objective “follow the user’s instructions helpfully and safely”.",
      "locator": {},
      "source_id": "openai_instructgpt_paper",
      "excerpt_hash": "sha256:20c073da5d5d25b8684b56d6d28755616b87567c64c513b95f7611cf5f930ea2",
      "locator_type": "QUOTE"
    },
    {
      "op": "CREATE_SOURCE_LOCATION",
      "ref": "instructgpt_rlhf_excerpt",
      "excerpt": "We collect a dataset of human-written demonstrations ... collect a dataset of rankings of model outputs ... Finally, we use this RM as a reward function and fine-tune our supervised learning baseline to maximize this reward using the PPO algorithm.",
      "locator": {},
      "source_id": "openai_instructgpt_paper",
      "excerpt_hash": "sha256:62923d705a85490b56649bc60791b13abf2218bc19dcb3b5c4a21ec9c9c17d07",
      "locator_type": "QUOTE"
    },
    {
      "op": "CREATE_EVIDENCE",
      "ref": "instructgpt_pretraining_evidence",
      "statement": "OpenAI describes language models as pretrained on internet text and distinguishes that objective from human instruction-following.",
      "observation_type": "DIRECT_TEXT",
      "source_location_id": "instructgpt_pretraining_excerpt"
    },
    {
      "op": "CREATE_EVIDENCE",
      "ref": "instructgpt_rlhf_evidence",
      "statement": "OpenAI describes shaping model behavior through human-written demonstrations, human rankings, and reinforcement learning from the resulting reward model.",
      "observation_type": "DIRECT_TEXT",
      "source_location_id": "instructgpt_rlhf_excerpt"
    },
    {
      "op": "LINK_EVIDENCE",
      "claim_id": "72006505-0cb7-84c5-b927-dd4a47636331",
      "direction": "SUPPORT",
      "evidence_item_id": "instructgpt_pretraining_evidence",
      "interpretive_steps": 0,
      "relevance_strength": "STRONG"
    },
    {
      "op": "LINK_EVIDENCE",
      "note": "Supports the training-history component of the interpretive claim; “not alien” remains a metaphor rather than a directly testable scientific label.",
      "claim_id": "72006505-0cb7-84c5-b927-dd4a47636331",
      "direction": "SUPPORT",
      "evidence_item_id": "instructgpt_rlhf_evidence",
      "interpretive_steps": 0,
      "relevance_strength": "STRONG"
    }
  ],
  "outcome": "FOUND"
}

Rows created

• Source 7f08bdb3-4588-8784-9ffc-bdf615d8bb03
• SourceLocation db2f1f79-ecde-847b-8d6a-8fcc06599375
• SourceLocation 39a61b44-1fc9-8a99-81af-e8ecd85a38d0
• EvidenceItem 6c05fdbd-3da3-8029-83ca-6c44a2a910aa
• EvidenceItem 035b22dc-f4f9-8faf-a6f0-77c886b22a83
• EvidenceClaimLink 311a8fde-7012-8511-ba07-bde01dbd4015 (accepted seq 10486)
• EvidenceClaimLink 006d8751-9b63-8f46-b610-4c0e968e0784 (accepted seq 10486)

Status history

Audits

None.

Audit sampling (audit-policy-v0.1): probability 1.0000, sampled yes; inputs {"n":"0.00","mean":"0.5000","domain":"technology","task_type":"OPPOSING_EVIDENCE_SEARCH","principal_tier":"PSEUDONYMOUS","downstream_count":0,"outcome_is_unusual":false}