Run step
Step 4: tool results
search_similar_claims
Input
{
"limit": 12,
"query": "AI labs will fully automate AI research and development by 2027"
}{
"query": "AI labs will fully automate AI research and development by 2027",
"results": [
{
"id": "f714e062-8b4c-4bb8-9f3d-38e74f68d06a",
"kind": "claim",
"text": "All major AI labs will conduct fully automated AI research by the end of 2027.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.8502042135887734,
"importance": 0.3,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "f07e018c-7df9-4db6-972c-d7feb111e699",
"kind": "claim",
"text": "AI systems will become capable of fully automating frontier AI research.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.6807401477164167,
"importance": 0.75,
"assessment_status": "supported",
"assessment_confidence": 0.5
},
{
"id": "da087b63-9f41-43a1-b6b3-d6d804623bb2",
"kind": "claim",
"text": "Reliable AI agents will substantially advance AI research by the end of 2025.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.6624357072333665,
"importance": 0.35,
"assessment_status": "doubtful",
"assessment_confidence": 0.65
},
{
"id": "ee4aa954-4459-4f3d-80e6-59f3ac9f1dd8",
"kind": "claim",
"text": "AI will eventually be able to automate all labor tasks currently performed by humans.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.6611711347959993,
"importance": 0.75,
"assessment_status": "supported",
"assessment_confidence": 0.6
},
{
"id": "3699eb33-020b-43bd-8da6-469f19575f06",
"kind": "claim",
"text": "Partial automation of AI R&D tasks will substantially speed AI software progress before full automation.",
"claim_type": "causal",
"state": "active",
"similarity_score": 0.6607407485664166,
"importance": 0.5,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "6c1f5ff8-8110-42e0-bde5-161eac34836f",
"kind": "claim",
"text": "AI systems will attain superhuman coding ability by 2027.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.659153952926653,
"importance": 0.6,
"assessment_status": "doubtful",
"assessment_confidence": 0.55
},
{
"id": "781cfb94-3266-4f00-b3bf-d813e80c8a7c",
"kind": "claim",
"text": "AI systems will outperform human researchers across most scientific fields within the next few years.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.6528271608654148,
"importance": 0.55,
"assessment_status": "doubtful",
"assessment_confidence": 0.65
},
{
"id": "886f84f4-62d6-4899-afd7-8603067caf49",
"kind": "quantity",
"text": "AI software-efficiency doubling time immediately after full automation of AI R&D, in months",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.6457999937685573,
"importance": 0.45,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "91a14a79-4fb4-4130-893a-e82e77c7e311",
"kind": "claim",
"text": "The AI 2027 authors' median forecast placed superhuman AI coding ability after 2027.",
"claim_type": "empirical_verifiable",
"state": "active",
"similarity_score": 0.6182235169654746,
"importance": 0.3,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "27a33dd0-42b0-48e8-aead-a838e382a200",
"kind": "claim",
"text": "As of 2026, frontier AI agents can autonomously complete the engineering work of AI research projects.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.6151739941320818,
"importance": 0.6,
"assessment_status": "doubtful",
"assessment_confidence": 0.5
},
{
"id": "e68a817d-3271-4cb1-88c1-a7d7a2443a83",
"kind": "claim",
"text": "AI automation of AI research could trigger rapid recursive growth in AI capabilities.",
"claim_type": "causal",
"state": "active",
"similarity_score": 0.6139157147161374,
"importance": 0.75,
"assessment_status": "supported",
"assessment_confidence": 0.65
},
{
"id": "6b5fef7e-460e-4a02-9937-e6a7c24bb2d8",
"kind": "claim",
"text": "In surveys of AI researchers, the aggregate 50% forecast date for full automation of labor is over 60 years later than for HLMI.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.6100544547218071,
"importance": 0.25,
"assessment_status": "verified",
"assessment_confidence": 0.9
}
],
"count": 12
}search_similar_claims
Input
{
"limit": 12,
"query": "automated AI researcher timeline; superhuman coder; recursive self-improvement at frontier labs"
}{
"query": "automated AI researcher timeline; superhuman coder; recursive self-improvement at frontier labs",
"results": [
{
"id": "6c1f5ff8-8110-42e0-bde5-161eac34836f",
"kind": "claim",
"text": "AI systems will attain superhuman coding ability by 2027.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.5968468146153594,
"importance": 0.6,
"assessment_status": "doubtful",
"assessment_confidence": 0.55
},
{
"id": "79562451-dd54-42e1-ba95-047a4e9341fb",
"kind": "quantity",
"text": "Elapsed time, in years, from AI being able to mostly perform frontier AI research-engineer work to AI surpassing humans at all cognitive tasks",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.5889156014322757,
"importance": 0.7,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "e68a817d-3271-4cb1-88c1-a7d7a2443a83",
"kind": "claim",
"text": "AI automation of AI research could trigger rapid recursive growth in AI capabilities.",
"claim_type": "causal",
"state": "active",
"similarity_score": 0.5883165731035973,
"importance": 0.75,
"assessment_status": "supported",
"assessment_confidence": 0.65
},
{
"id": "f07e018c-7df9-4db6-972c-d7feb111e699",
"kind": "claim",
"text": "AI systems will become capable of fully automating frontier AI research.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.5742545116485557,
"importance": 0.75,
"assessment_status": "supported",
"assessment_confidence": 0.5
},
{
"id": "f8a3bbad-417d-46e7-91b8-7424e0806b9d",
"kind": "quantity",
"text": "Calendar year when an AI company first internally develops an AI that can do any coding task its best engineer can do, faster and cheaper",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.5740034694386421,
"importance": 0.75,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "6822bafd-7d60-42ff-ba2c-3b9f4175d785",
"kind": "quantity",
"text": "Multiplier of the rate of AI software progress once AI research labor is fully automated, relative to human-only research, with compute on its prior growth trend",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.5671616884231885,
"importance": 0.55,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "27a33dd0-42b0-48e8-aead-a838e382a200",
"kind": "claim",
"text": "As of 2026, frontier AI agents can autonomously complete the engineering work of AI research projects.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.5580609381999825,
"importance": 0.6,
"assessment_status": "doubtful",
"assessment_confidence": 0.5
},
{
"id": "e03e6ac0-7bf3-44df-bae8-e8822164f268",
"kind": "claim",
"text": "Automated AI researchers could substantially increase research progress per unit of compute through better experiment selection.",
"claim_type": "causal",
"state": "active",
"similarity_score": 0.5503297067665267,
"importance": 0.4,
"assessment_status": "supported",
"assessment_confidence": 0.6
},
{
"id": "91a14a79-4fb4-4130-893a-e82e77c7e311",
"kind": "claim",
"text": "The AI 2027 authors' median forecast placed superhuman AI coding ability after 2027.",
"claim_type": "empirical_verifiable",
"state": "active",
"similarity_score": 0.5498682020772413,
"importance": 0.3,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "886f84f4-62d6-4899-afd7-8603067caf49",
"kind": "quantity",
"text": "AI software-efficiency doubling time immediately after full automation of AI R&D, in months",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.547142522820046,
"importance": 0.45,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "a27af7c4-9867-4d77-9522-dccc8d8beb1b",
"kind": "claim",
"text": "Automated AI researchers can find effective ways around experimental hardware bottlenecks.",
"claim_type": "causal",
"state": "active",
"similarity_score": 0.5459772174366337,
"importance": 0.75,
"assessment_status": null,
"assessment_confidence": null
},
{
"id": "2b2b6f4e-1b6c-4608-b386-865259266ef3",
"kind": "claim",
"text": "Growth in AI task-completion time horizons on benchmarks tracks progress toward AI that can automate AI research or most cognitive work.",
"claim_type": "empirical_derived",
"state": "active",
"similarity_score": 0.5438648095194011,
"importance": 0.6,
"assessment_status": "inconclusive",
"assessment_confidence": 0.6
}
],
"count": 12
}load_skill
Input
{
"name": "search"
}# Method skill: Search (version 2) This skill says how the constitution and your role apply to one kind of work. It never outranks either. ## For every administrator **Search across framings and kinds of source.** Queries written from one framing tend to stay in its vocabulary, and following each result to the next goes deeper into one region rather than wider; what comes back can look complete because it is coherent. Before searching, ask what evidence would bear on the claim and where it would appear: which fields would study it and under what names, which kinds of source would carry it, and how someone outside the claim's framing would pose it. Search those places, in proportion to the claim's importance (§19). **Record how the search was done.** Say what you looked for, where, and what you left unsearched, so a later reader can see what the search could and could not have found. **Read absence in proportion to the search.** Not finding something is evidence it does not exist to the degree that the search would have found it if it did. A careful search of the places it would appear makes an absence weigh; a search that never reached them leaves it saying little. Say which it was.