“Alignment is still shaky; you still wouldn’t be able to use it reliably to guide robots or scientific discovery, the kinds of things that made me excited about A(G)I in the first place.”
Our reading, the sentence that was checked:
GPT-4's alignment remains shaky and it cannot be used reliably to guide robots or scientific discovery.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, GPT-4’s successes, and GPT-4’s failures, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-gpt-4s-successes-and-gpt-4s-failures-d8a265-a1
We recorded 12 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2023-03-15. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- 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)
From the speaker
- Alignment is still shaky
- still wouldn’t be able to use it reliably
- guide robots or scientific discovery
Our interpretation
- Still describes the original model at publication, rather than forecasting that later systems will never work.
- Shaky alignment means meaningful failures to follow intended behavioral constraints.
- Reliable guidance means repeatable, dependable performance on relevant robotic or scientific workflows. A successful demonstration alone does not establish reliability.
- The wording does not specify whether GPT-4 acts alone, with tools, or under human supervision; no single configuration is silently assumed.
- The launch-to-post interval is the latest publicly documented snapshot of this ongoing state.
Counts as supported if: Launch-model evidence establishes remaining alignment failures and inadequate reliability on relevant robot-control and scientific-discovery workflows.
Counts as contradicted if: Representative evaluations establish dependable guidance of the relevant workflows by the original GPT-4, or establish that the alleged alignment failures were resolved.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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 alignment clause is supported by documented launch-model jailbreak failures. Those failures do not establish that GPT-4 could not reliably guide robots or scientific discovery in relevant configurations. Later scientific automation supplies useful countercontext, but it does not settle what was dependable at publication. The full claim remains unresolved.
Limitations
- Neither missing launch-period demonstrations nor general language-model errors prove application-wide inability.
- Later successes are context, not retrospective proof of launch-period reliability.
Evidence (2 items)
Supports the claim: OpenAI,
Training on this data improves model robustness but does not fully solve the problem of “jailbreaks” leading to harmful content.
Why it matters here: The passage establishes remaining alignment failures in GPT-4-launch despite improved robustness. The event date denotes the model's launch state; this evidence does not directly measure robot-control or scientific-workflow reliability.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Nature,
Our findings demonstrate the versatility, efficacy and explainability of artificial intelligence systems like Coscientist in advancing research.
Why it matters here: The paper reports a GPT-4-based system that uses tools to plan and execute chemical experiments, including reaction optimization. Its experiments are not dated to the target interval, so it challenges an unrestricted impossibility reading but cannot contradict this launch-period observation.
- Where
- Abstract
- Publisher
- Nature
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
Direct evaluations attributable to the original model and target period, with defined tasks, supervision, repeated trials and meaningful reliability thresholds for robotics and scientific workflows.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, GPT-4’s successes, and GPT-4’s failures
Alignment is still shaky; you still wouldn’t be able to use it reliably to guide robots or scientific discovery, the kinds of things that made me excited about A(G)I in the first place.
- Where
- paragraph 4
- Published
- 2023-03-15
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-d8a265-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-gpt-4s-successes-and-gpt-4s-failures-d8a265-a1.