ClaimsSam Altman

“not until 2020 did we train GPT-3, which was the first model that I think actually was over some threshold of usability.”

Sam Altman, OpenAI CEO 4 September 2026 Transcript: Sam Altman Interview on OpenAI, New AI Model and Global Growth - G20 Innovation Summit (Singju Post) archived copy

Factual AI capabilities

OpenAI trained GPT-3 in 2020, which Altman considers the first model above a threshold of usability.

Mostly true True in substance, wrong in a detail.

Factual: Statements about the present or the past, checked against the record: did they hold up?

The second judge would not resolve this claim; the verdict shown is the first judge's.

Said for an organisation they lead or work for; it counts on both records.

What would count as true or false

The factual part is when GPT-3 was trained; Altman's view on usability is opinion and is accepted as stated. True: records show GPT-3 was trained (or its training finished) in 2020 and it was first released or announced in 2020. Mostly true: training mainly took place in late 2019 but finished and was announced in 2020, so '2020' is off by a minor detail. Mostly false: GPT-3 was clearly trained well before 2020 and only announced in 2020. False: GPT-3 was neither trained nor released in 2020.

Set before the research began.

Evidence (11)

  1. E1 GPT-3 - Wikipedia Wikipedia · 31 July 2020

    “On May 28, 2020, an arXiv preprint by a group of 31 engineers and researchers at OpenAI described the achievement and development of GPT-3”

  2. E2 GPT-3 - Wikipedia Wikipedia · 31 July 2020

    “Release | May 29, 2020 (publication); June 11, 2020 (OA API beta)”

  3. E3 GPT-3 175B (davinci) Epoch AI

    “GPT-3 175B (davinci) is a closed-weights model from OpenAI, released May 28, 2020.”

  4. E4 OpenAI API OpenAI · 11 June 2020 Own record

    “We’re releasing an API for accessing new AI models developed by OpenAI.”

  5. E5 GPT-3 - Wikipedia Wikipedia · 31 July 2020

    “Release | May 29, 2020 (publication); June 11, 2020 (OA API beta)”

  6. E6 GPT-3 - Wikipedia Wikipedia · 31 July 2020

    “Generative Pre-trained Transformer 3 (GPT-3) was a large language model released by OpenAI in 2020 as part of the company's GPT series of models.”

  7. E7 GPT-3 - Wikipedia Wikipedia · 31 July 2020

    “On May 28, 2020, an arXiv preprint by a group of 31 engineers and researchers at OpenAI described the achievement and development of GPT-3”

  8. E8 GPT-3 - Wikipedia Wikipedia · 31 July 2020

    “Lambdalabs estimated a hypothetical cost of around $4.6 million US dollars and 355 years to train GPT-3 on a single GPU in 2020”

  9. E9 gpt-3/model-card.md at master · openai/gpt-3 GitHub (openai/gpt-3 repository) · September 2020 Own record

    “The May 2020 version of GPT-3 was trained on a dataset created in November 2019, so has not been trained on any data more recent than that.”

  10. E10 Microsoft supercomputers Glenn K. Lockwood (personal site) · 28 October 2026

    “The first AI supercomputer we built back in 2019 for OpenAI was used to train GPT-3 and the size of that infrastructure”

  11. E11 gpt-o1-pro-on-time-to-product - HackMD HackMD (user-published document, author unknown)

    “the GPT-3 training run was reportedly completed in early 2020, soon after which the paper was published”

Reasoning

Independent sources date GPT-3's public release to 2020: the arXiv preprint on May 28, 2020, followed by the API beta in June 2020 (E1, E2, E3). OpenAI's model card says the May 2020 version was trained on a dataset created in November 2019, so training cannot have taken place well before 2020 (E9). Only a weak, unsourced document says the training run was completed in early 2020 (E11), and E10 mentions a 2019 supercomputer used for training without giving training dates. The evidence is consistent with training in late 2019 and early 2020 and release in 2020, but the exact training dates are not firmly documented, so the '2020' claim holds with a possible minor imprecision.

Cite this claim

Sam Altman, “OpenAI trained GPT-3 in 2020, which Altman considers the first model above a threshold of usability.” (said 4 September 2026). Verdict: mostly true. Reliability API, claim c_3c879a3d97f2, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_3c879a3d97f2

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

For Sam Altman, or someone speaking for them: a correction or counter-evidence. Replies are read before anything is shown; an approved reply shows your name, role, date, statement and link. Your email address is never shown and is deleted 30 days after the decision (privacy).

More factual claims by Sam Altman

  1. Resolved true Factual 4 September 2026
    “that it is so capable that it is discovering new knowledge, doing science, able to add huge economic value, create entire pieces of complex software.”

    Altman said that a system "we now have" is capable of discovering new knowledge, doing science, adding huge economic value and creating entire pieces of complex software.

  2. Mixed Factual 11 September 2025
    “They now hallucinate a little bit.”

    Current AI models now hallucinate only a little bit.

  3. Mixed Factual 11 September 2025
    “They don’t do anything unless you ask, right?”

    AI models such as ChatGPT do not do anything unless a user asks them to.

  4. Mostly true Factual 15 August 2025
    “But in terms of figuring out what questions to ask or to work on something over a very long period of time, we are definitely not close to human performance.”

    GPT-5 is not close to human performance at figuring out what questions to ask or working on a task over a very long period.

  5. Mostly true Factual 14 August 2025
    “It’s an integrated model, so you don’t have to pick in our model switcher and know if you should use GPT-4o or o3 or o4 mini or any of the complicated things.”

    GPT-5 is an integrated model, so users do not need to choose between GPT-4o, o3 or o4 mini in the model switcher.

All of them

Verdicts are made automatically; how verdicts are made.