Breaking: OpenAI's efforts at pure scaling have hit a wall.
We recorded 9 claims from this post. 2 were checked against evidence. 7 can’t be checked. 6 more are held. Read the post at its source.
Claims from this post (9)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: UnresolvedSaid
Forecast
“For nearly three years I have been saying that the pure scaling of LLMs – adding more data and more parameters – would eventually run out, and that it would fail to solve hallucinations and boneheaded errors and that scaling laws were merely empirical generalizations, rather than physical laws.”
Our reading: Marcus restates his long-held prediction that pure scaling of LLMs (adding more data and more parameters) would eventually run out and would fail to solve hallucinations and boneheaded errors, and that scaling laws are merely empirical generalizations rather than physical laws.
No deadline stated Evidence as of
Limitation: Available model comparisons change several factors simultaneously.
Assessment: Can’t be checkedSaid
Observation
“All this time, it turns out, Altman was bluffing.”
Our reading: Sam Altman's comments implying progress toward AGI were a bluff.
No horizon
Assessment: Can’t be checkedSaid
Conditional forecast
“If they called Orion GPT-5, reactions would likely converge for many on “is that all there is?”. OpenAI’s mystique might well take a serious hit.”
Our reading: If OpenAI called Orion GPT-5, many would likely react with disappointment and OpenAI's mystique might take a serious hit.
No horizon
Only if: if Orion were called GPT-5
Assessment: Can’t be checkedSaid
Observation
“The company’s future rests on them doing something truly revolutionary, and Orion turns out not to be that.”
Our reading: OpenAI's future depends on something revolutionary, and Orion is not revolutionary.
No horizon
Assessment: SupportedSaid
Forecast
“Second, pure scaling will no longer be the means of attack.”
Our reading: OpenAI will no longer rely on pure scaling as its means of building GPT-5.
Settled before its deadline Evidence as of
Limitation: The disclosure establishes the additional mechanisms, not their precise share of capability gains.
Assessment: Can’t be checkedSaid
Observation
“If Altman knew how to build AGI, he would build it. Instead, he rechristened Orion as GPT 4.5 rather than 5, which likely means that despite major bills, Orion never met the expectations so-called scaling laws had implied.”
Our reading: Altman does not know how to build AGI, and Orion likely never met the expectations implied by scaling laws despite major costs.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“Neurosymbolic AI, which I have long championed, may turn out to be part of the mix.”
Our reading: Neurosymbolic AI may be part of the eventual route to AGI.
No horizon
Assessment: Can’t be checkedSaid
Forecast
“In all likelihood AGI will some day arrive, with or without LLMs at the core, and likely with all kinds of other gadgets included.”
Our reading: AGI will likely arrive some day, likely incorporating many components beyond LLMs.
No horizon
Qualifier: with or without LLMs at the core
Assessment: Can’t be checkedSaid
Observation
“Scaling is useful, but it is not now and never was the whole answer.”
Our reading: Scaling is useful but is not, and never was, sufficient on its own to reach AGI.
No horizon Evidence as of
Limitation: The case cites no evidence items, so this record does not establish whether AGI has been reached or how.
Held from this post (6)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The two reviews reached different assessments. Our reading, not assessed: No one, including Altman, has yet figured out how to get to AGI.
- The two reviews read the claim differently
- The criteria are ambiguous. Our reading, not assessed: OpenAI will throw the kitchen sink at future efforts to build GPT-5, including test-time compute and, Marcus suspects, massive amounts of synthetic data, which appears to work better for semi-closed domains like math and coding than in the open-ended real world.
- The record does not meet the current rules. Our reading, not assessed: Orion, OpenAI's last great effort at pure scaling, has failed, and Marcus infers that OpenAI wants to ship it now because they can get no further with it.
- The two reviews read the claim differently
- The two reviews read the claim differently