“Meanwhile, large language models themselves are quickly becoming commodities; it’s not clear that there is any unique, protectable intellectual property here.”
Our reading:
LLMs are quickly becoming commodities, and it is unclear OpenAI has any unique, protectable IP.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
The contemporary record supports the direction Marcus described: OPT and BLOOM offered competing capabilities, and OpenAI cut standard-model prices. But it does not establish how quickly practical substitution was spreading across ordinary uses. HELM documented both substantial improvement and remaining adaptation differences; OPT also restricted commercial use. These are limits on the evidence, not proof that commoditization was absent. The qualified IP remark cannot fairly be converted into a claim that OpenAI owned no IP. The combined statement remains unresolved under its ordinary commercial reading.
Caveats
- The period and meaning of rapid commoditization require interpretation.
- Model papers report technical comparisons rather than customer substitution.
- Later model releases cannot settle this January 2023 observation.
Evidence (5 items)
Supports the claim: Meta researchers / arXiv,
comparable to GPT-3
Why it matters: The authors report a technically comparable alternative to GPT-3, supporting the direction of commoditization. This comparison does not establish commercial interchangeability or parity with every OpenAI offering.
- Where
- Abstract
- Publisher
- Meta researchers / arXiv
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
Supports the claim: BigScience researchers / arXiv,
competitive performance
Why it matters: Another research group reported competitive performance across multiple benchmarks. This supports increasing technical alternatives, but does not measure how rapidly buyers could substitute between suppliers.
- Where
- Abstract
- Publisher
- BigScience researchers / arXiv
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
Supports the claim: OpenAI,
September 1, 2022
Why it matters: The accompanying table reduces standard Davinci pricing from $0.06 to $0.02 per thousand tokens and also lowers other standard-model prices. This supports falling access costs, although one supplier's reductions do not establish market-wide substitutability.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
Context only: Stanford Center for Research on Foundation Models,
not public knowledge
Why it matters: HELM describes substantial open-model improvement alongside performance and prompting differences, and identifies undisclosed training methods. These findings limit inferences about practical substitution and IP, without disproving an ongoing commoditization trend.
- Publisher
- Stanford Center for Research on Foundation Models
- Kind of source
- Independent measurement
- About events on
- Published
Context only: Meta,
noncommercial license
Why it matters: OPT's noncommercial restriction limits its relevance as a substitute for commercial OpenAI services. Its technical comparability remains evidence of reproducible capabilities.
- Publisher
- Meta
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Where we had to interpret
- The plain reading is a rapid movement toward competing LLMs serving similar needs with less dependence on a particular supplier. It does not require complete interchangeability or the disappearance of frontier performance differences.
- Use 2022, the most recent completed calendar year before the post, to assess the ongoing industry trend. This observation does not promise an outcome during 2023.
- Converging task performance, widening practical availability and declining comparable prices are indicators of commoditization. Model counts alone are insufficient; documented customer switching is helpful but not required.
- Read the IP clause as uncertainty about an exclusive technological advantage that competitors cannot readily reproduce. It is not an assertion that OpenAI owns no intellectual property.
Right if: Evidence from the target period establishes a substantial movement toward practical substitutes across ordinary LLM uses, through improving comparable capabilities and access or pricing. The qualified IP clause requires assessing what was publicly established about an exclusive, protectable advantage, without demanding proof that no IP existed.
Wrong if: Evidence establishes that practical substitutability was stagnant or declining across ordinary uses during the target period. Alternatively, contemporaneous public records clearly establish the exclusive, protectable technological advantage whose existence Marcus questioned. A remaining performance lead or an isolated patent is insufficient by itself.
We wrote this test after the outcome was known.
What would settle it
Comparable measurements across 2022 showing how rapidly multiple accessible suppliers became practical substitutes for common tasks, including adaptation costs and pricing, plus contemporaneous evidence clarifying any claimed exclusive technological protection.
We found 7 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Is Microsoft about to get the deal of the century? Or is Sam Altman unloading OpenAI at just the right time?, paragraph 36.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-29. The post may have been edited after 2023-01-11.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Is Microsoft about to get the deal of the century? Or is Sam Altman unloading OpenAI at just the right time?
Meanwhile, large language models themselves are quickly becoming commodities; it’s not clear that there is any unique, protectable intellectual property here.
- Where
- paragraph 36
- Published
- 2023-01-11
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
A reviewer said our test wasn’t fair to what was actually said.