“Its natural language output still won’t be something that one can reliably hook up to downstream programs; it won’t be something, for example, that you can simply and directly hook up to a database or virtual assistant, with predictable results.”
Our reading, the sentence that was checked:
GPT-4's natural language output still won't be something one can reliably hook up to downstream programs, for example simply and directly to a database or virtual assistant with predictable results.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What to Expect When You’re Expecting … GPT-4, prediction 4 of 7.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-what-to-expect-when-youre-expecting-238701-a1
We recorded 27 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 2022-12-25. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
Conditions
- natural language output (scope, stated by the speaker)
- simply and directly (qualifier, stated by the speaker)
From the speaker
- Its natural language output
- reliably hook up to downstream programs
- simply and directly
- a database or virtual assistant
- with predictable results
Our interpretation
- The target is GPT-4 and its documented updates, rather than every later OpenAI model.
- The plain reading concerns dependable automatic execution of intended operations, not merely syntactically valid output.
- Simply and directly allows ordinary interface code, but not substantial semantic correction or human review of each operation.
- Natural language output and simply and directly restrict the scope; they are not conditional antecedents.
- Reliability means practical dependability across representative requests, not perfect performance on every possible input.
Counts as supported if: Representative integration evaluations show that directly consumed GPT-4 output causes sufficiently frequent invalid or unintended operations to prevent dependable routine use.
Counts as contradicted if: A documented GPT-4 integration directly converts representative user requests into correct downstream operations with measured practical reliability and without substantial semantic correction.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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 deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The evidence supports the concern that output formatting and semantic correctness require care. It also shows that GPT-4 gained interfaces expressly intended for downstream programs. Neither the complex-schema result nor the interface announcements establish practical reliability under the claim's direct-connection restriction. With no deadline and no decisive matching evaluation, the forecast remains unresolved.
Limitations
- The numerical evaluation is provider-reported and concerns complex schemas.
- Constrained decoding changes the integration method.
- A formatting guarantee does not establish that the intended operation is correct.
Evidence (3 items)
Context only: OpenAI,
gpt-4-0613 scores less than 40%.
Why it matters here: This is evidence of substantial difficulty following complex output schemas. It measures schema compliance rather than correct execution of ordinary database or assistant requests, so it does not by itself settle the forecast.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: OpenAI,
more reliably get structured data
Why it matters here: GPT-4-0613 introduced structured arguments for external functions, including database and assistant examples. This challenges a broad claim that integration is impossible, but supplies no measured operational reliability and uses a structured interface.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: OpenAI,
model mistakes
Why it matters here: The release extends constrained function outputs to GPT-4-0613 and later tool-supporting models, while warning that values can still be wrong. It establishes progress in formatting without establishing dependable semantic behavior.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A representative, reproducible GPT-4 integration evaluation reporting correct operations, errors, retries, and intervention under a clearly described direct-connection design and practical acceptance criterion.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What to Expect When You’re Expecting … GPT-4
Its natural language output still won’t be something that one can reliably hook up to downstream programs; it won’t be something, for example, that you can simply and directly hook up to a database or virtual assistant, with predictable results.
- Where
- prediction 4 of 7
- Published
- 2022-12-25
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-to-expect-when-youre-expecting-238701-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-what-to-expect-when-youre-expecting-238701-a1.