ClaimsEliezer Yudkowsky

“Beating GPT-4 on GI just forces OpenAI to release 4.5.”

Eliezer Yudkowsky, writing on X 5 March 2024 Post by @ESYudkowsky on X, 2024-03-05 (X)

Prediction AI capabilities

If a rival AI developer beats GPT-4 on general intelligence, OpenAI will be forced to release GPT-4.5.

Mixed Partly true and partly false in roughly equal measure.

Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?

What would count as true or false

This is a conditional prediction. The condition is met when a rival developer releases a model widely judged (by benchmark leaderboards or broad expert or press consensus) to be better than GPT-4 at general capability. True: the condition is met and OpenAI then releases a model branded GPT-4.5, with the release reported or reasonably presented as a response to losing the lead. Mostly true: OpenAI answers the rival with a clearly upgraded GPT-4 successor under a different name, or releases GPT-4.5 but the causal link to the rival is only weakly shown. Mixed: GPT-4.5 is released eventually, but long after the condition was met, with no evident link to competitive pressure. Mostly false: OpenAI releases no GPT-4.5 and no prompt competitive successor for a long period while rivals stay ahead. False: rivals clearly surpass GPT-4, and OpenAI never releases a GPT-4.5 and makes no competitive release in response. If no rival ever surpasses GPT-4, the claim is unresolvable.

Set before the research began.

Evidence (4)

  1. E1 The AI wars heat up with Claude 3, claimed to have “near-human” abilities Ars Technica · 4 March 2024

    “According to Anthropic, Claude 3 Opus beats GPT-4 on 10 AI benchmarks, including MMLU (undergraduate level knowledge), GSM8K (grade school math), HumanEval (coding), and the colorfully named HellaSwag (common knowledge).”

  2. E2 Simon Willison on gpt-4 Simon Willison's Weblog

    ““The king is dead”—Claude 3 surpasses GPT-4 on Chatbot Arena for the first time.”

  3. E3 Introducing GPT-4.5 OpenAI · 27 February 2025

    “GPT‑4.5 is an example of scaling unsupervised learning by scaling up compute and data, along with architecture and optimization innovations.”

  4. E4 OpenAI launches GPT-4.5, its largest model to date | TechCrunch TechCrunch · 1 March 2025

    “GPT-4.5 is OpenAI’s largest model to date, trained using more computing power and data than any of the company’s previous releases.”

Reasoning

The condition appears to have been met: Ars Technica reported that Claude 3 Opus beat GPT-4 on 10 benchmarks (E1), and Simon Willison's weblog noted that Claude 3 surpassed GPT-4 on Chatbot Arena (E2). OpenAI released a model branded GPT-4.5 on 2025-02-27 (E3, E4), nearly a year after the condition was met. The quoted passages from OpenAI's announcement and TechCrunch's coverage describe GPT-4.5 in terms of scaling compute and data and do not mention a rival model or a competitive response (E3, E4). That matches the criteria's "mixed" case: an eventual release with no evident link to competitive pressure.

Cite this claim

Eliezer Yudkowsky, “If a rival AI developer beats GPT-4 on general intelligence, OpenAI will be forced to release GPT-4.5.” (said 5 March 2024). Verdict: mixed. Reliability API, claim c_51a36d84a2e5, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_51a36d84a2e5

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

Reply to this claim

For Eliezer Yudkowsky, 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 prediction claims by Eliezer Yudkowsky

  1. Mostly true Prediction February 2023
    “But I do feel quite confident that the upper range of that untapped potential is insufficient to outsmart all the living humans and implement the scenario that I'm worried about.”

    Yudkowsky said he felt quite confident that the upper range of the untapped potential of ChatGPT-style systems is insufficient to outsmart all living humans and carry out the catastrophe scenario he worries about.

  2. Resolved false Prediction 6 April 2023
    “The current crop of people will not try it.”

    The current AI developers will not try the alignment approach he described (filtering training data to nice examples and training on it).

  3. Mostly false Prediction 6 April 2023
    “So probably Paul’s going to win, but slight moral victory.”

    Paul Christiano will probably win his forecast disagreement with Yudkowsky on an AI getting IMO gold by 2025, with prediction-market odds then running around 30%.

  4. Resolved true Prediction 6 April 2023
    “Alignment will not be solved in a few years.”

    AI alignment will not be solved within a few years.

  5. Resolved true Prediction 6 April 2023
    “even if GPT-5 doesn’t end the world, which I agree is like more than 50% of where my probability mass lies”

    GPT-5 will not end the world, with more than 50% of Yudkowsky's probability on that outcome.

All of them

Verdicts are made automatically; how verdicts are made.