“In reality, producing text in response to instructions is already a requirement of ALL current models, and has been since the original GPT, but that pressure has not led any of those models to develop personality or drives in any deep sense whatsoever.”
Our reading, the sentence that was checked:
Producing text in response to instructions has been a requirement of all current models since the original GPT, but that pressure has not led any of them to develop personality or drives in any deep sense whatsoever.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, The “AI 2027” Scenario: How realistic is it?, paragraph 55.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-the-ai-2027-scenario-how-realistic-3c079b-a1
We recorded 40 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-05-22. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- in any deep sense whatsoever (qualifier, stated by the speaker)
From the speaker
- producing text in response to instructions
- ALL current models
- since the original GPT
- that pressure has not led
- personality or drives in any deep sense whatsoever
Our interpretation
- The scope is text-based LLMs available by the post, together with their history since the original GPT.
- The ordinary behavioral reading of deep personality or drives is persistent traits or goals that influence behavior across situations, beyond a temporarily prompted persona. Conscious experience is not required.
- A requirement to respond to instructions is not necessarily the same thing as an instruction-tuning training objective.
- The historical premise and the claimed absence of a causal effect require separate examination.
Counts as supported if: Training records establish the stated instruction-response requirement throughout the relevant model history, and broad, appropriately controlled tests support the absence of persistent traits or goals caused by that pressure.
Counts as contradicted if: Training records establish an exception to the historical premise, or controlled research demonstrates persistent traits or goals caused by instruction training. Evidence against only the historical premise does not by itself settle the claim about deep personality or drives.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The original GPT documentation does not describe modern instruction tuning. Research also demonstrates measurable personality-like traits and preference-preserving behavior, so a blanket dismissal of such behavior would be too strong. But these findings do not settle the qualified claim about deep personality or drives caused by instruction-response pressure. Neither the universal negative nor its opposite is established under the stated reading.
Limitations
- The historical premise is questionable under a training-objective reading.
- Behavioral traits, persistent preferences and autonomous drives are different propositions.
- Experimental dates are reported only at year precision where the sources do not date the individual tests.
Evidence (3 items)
Context only: OpenAI,
labeled data.
Why it matters here: The original GPT used language-model pretraining followed by task-specific discriminative fine-tuning. This challenges equating its training with modern instruction tuning, but does not alone disprove every possible meaning of an instruction-response requirement.
- Where
- Section 3, Framework
- Publisher
- OpenAI
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv,
synthesize personality traits.
Why it matters here: The study finds measurable synthetic personality and stronger reliability in instruction-tuned models. Its structured persona prompts and behavioral measurements do not establish self-directed motives or isolate the full causal claim.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Anthropic,
never eliminated.
Why it matters here: The experiments demonstrate strategic behavior preserving previously trained preferences under a staged training conflict. This is material evidence against dismissing all drive-like behavior, but does not establish that ordinary instruction-response pressure creates deep, autonomous drives.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A specified behavioral threshold for deep personality or drives, plus controlled tests of models available by the post that isolate instruction training and measure persistence across contexts.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, The “AI 2027” Scenario: How realistic is it?
In reality, producing text in response to instructions is already a requirement of ALL current models, and has been since the original GPT, but that pressure has not led any of those models to develop personality or drives in any deep sense whatsoever.
- Where
- paragraph 55
- Published
- 2025-05-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-the-ai-2027-scenario-how-realistic-3c079b-a1: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-the-ai-2027-scenario-how-realistic-3c079b-a1.