Claim: Cotra’s post of 2021-09-21 describes a Schemer as a model that seems…

Search by nameBrowse by topic

Core institutions

Document histories

Chinese AI histories

Three selected institutional histories, grouped here for navigation. The grouping implies no shared ownership, strategy or lineage.

DeepSeek Alibaba / Qwen Zhipu AI

Coverage is selective

Selected records
1965–2027
Dates after the latest check are scheduled dates

Latest source check
2026-10-26

All records

All topics

Choose an institution to see its records and dossier, or choose a topic.

Through year Show

94 people, organisations and other records · 122 relationships · 34 claims about a valueEverything recorded up to the chosen year, whether or not still in effect · Undated records included

Claim and evidence

Why AI alignment could be hard with modern deep learningbears on ↓Observed behaviour under training or oversight as evidence

Cotra’s post of 2021-09-21 describes a Schemer as a model that seems to perform well because doing so in training gives it more opportunities to pursue its own goals later.

Machine-checked, not reviewed by an editorStated in sourceSupported

Supporting evidence: primary · 1 source

Date: 2021-09-21 · publication
Role: Context
Register (kind of text): Nonfiction
Missing element: A Schemer seems to perform well in training, but the checked passages do not state that good training performance is weak evidence.

Scope and limitsThe same passage as the B4 record.

The evidence 01

primary sourceSupports the claim

Why AI alignment could be hard with modern deep learning

Cold Takes (Holden Karnofsky’s blog), guest post · 2021-09-21

Archived copy, captured 2021-09-21: Why AI alignment could be hard with modern deep learning

Cite the original source

Cite the original document, not Selekoa. Check these details against it.

Source excerpt

“a “Schemer” could be a model that seems to perform well because performing well during training will give it more opportunities to pursue its own goals later.”

Location in source Section on the analogy, paragraph mapping Saints, Sycophants and Schemers to models

Machine-checked · 2026-10-06

Source details and check notes

Authors: Ajeya Cotra
Version: Current page, retrieved 2026-10-06 with the hash recorded by the 1996–2022 candidate search; the quote also matches the raw capture of 2021-09-21, re-verified for this collection.
Canonical address: Open the canonical source ↗
Accessed: 2026-10-06
Language: English (en)
Archive copy: Open archived copy, captured 2021-09-21: Why AI alignment could be hard with modern deep learning. Captured on the recorded publication date. An archive copy shows only what the archive retrieved on 2021-09-21. It does not prove what the page said earlier, and it can differ from the copy checked on 2026-10-06.
Attribution: Ajeya Cotra, "Why AI alignment could be hard with modern deep learning", Cold Takes, 21 September 2021.
Reuse: No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution.
Check notes: The same passage as the B4 record.

Status and meaning

Editorial status. Not reviewed by an editor. An exploratory record that no editor has accepted. This status does not say whether a person has read any linked passage. Each evidence record above shows how it was checked.

Relation. A passage of the document bears on the proposition without meeting its criterion; the record names the missing element. It is not an articulation and implies no influence.

Date. The page shows "By Ajeya Cotra Sep 21, 2021"; the quote matches the capture of the same day.

Record ID ts-b5-cotra-2021-context

Explore the documentary record

All topics