“Companies like OpenAI that have talked big about investing in safety research and then stinted on their promises”
Our reading, the sentence that was checked:
OpenAI talked big about investing in safety research and then failed to deliver on those promises.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why California’s AI Safety Bill should (still) be signed into law - and why that won’t be nearly enough, paragraph 14.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-why-californias-ai-safety-bill-should-270f2b-a2
We recorded 21 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 2024-08-20. 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
- Companies like OpenAI
- investing in safety research
- stinted on their promises
Our interpretation
- The plain reading is that OpenAI materially underprovided resources it had publicly promised for safety research by the post date.
- The Superalignment commitment is a plausible instance to investigate, but Marcus does not identify that promise or a particular investment measure.
- Assess the historical sequence of public commitments and subsequent delivery through August 20, 2024. Do not treat an unfinished multiyear commitment as already breached merely because its full allocation had not been delivered.
Counts as supported if: An identifiable public safety-research resource commitment and comparable delivery records establish a material shortfall against what was due by the post date.
Counts as contradicted if: Comparable records show that OpenAI delivered the identified commitments due by the post date, without the alleged material shortfall.
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
On July 5, 2023, OpenAI committed 20% of its then-secured compute to superintelligence alignment over four years, without specifying interim allocations. [e1] On May 17, 2024, Jan Leike reported compute shortages affecting his team during the preceding months. [e2] WIRED reported that day that the team had disbanded and its work would be absorbed into other OpenAI research efforts. [e3] These sources document a promise and reported shortages, but do not establish a material shortfall against resources due by August 20, 2024. [e1] [e2] [e3]
Limitations
- The packet does not establish which particular promise Marcus meant.
- No comparable allocation records show what was due and delivered by August 20, 2024. [e1] [e2] [e3]
- Team-level shortages and dissolution do not measure delivery across the other teams contributing to the four-year commitment. [e1] [e2] [e3]
Evidence (3 items)
Context only: OpenAI,
20%
Why it matters here: This establishes a substantial public commitment with a four-year delivery window. It does not specify a minimum allocation due during the first year.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Measures
- OpenAI, Share of compute secured by July 5, 2023 committed to superintelligence alignment over four years percent of secured compute, company reported, 2023-07-05 to 2027-07-05
- Describes events of
- Public from
Context only: Jan Leike, personal account,
struggling for compute
Why it matters here: The departing team leader directly reported resource shortages. This is personal insider testimony, not an official company statement or a measurement of the pledged allocation.
- Publisher
- Jan Leike, personal account
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Measures
- OpenAI, Qualitative availability of compute to the Superalignment team, estimate, ? to 2024-05
- Describes events of
- Public from
Context only: WIRED,
other research efforts
Why it matters here: WIRED reported company confirmation that the team was dissolved and its work reassigned. Dissolution therefore does not by itself establish abandonment of the broader commitment.
- Publisher
- WIRED
- Kind of source
- News report
- Describes events of
- Public from
What would settle it
Identify the relevant public safety-research commitment, its resource baseline, beneficiaries, and delivery schedule. Comparable records of resources due and actually delivered by August 20, 2024 would establish whether a material shortfall occurred.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why California’s AI Safety Bill should (still) be signed into law - and why that won’t be nearly enough
Companies like OpenAI that have talked big about investing in safety research and then stinted on their promises
- Where
- paragraph 14
- Published
- 2024-08-20
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-why-californias-ai-safety-bill-should-270f2b-a2: 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-why-californias-ai-safety-bill-should-270f2b-a2.