Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
79 72 85 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
80 72 88 -
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 1 promise resolved, 5 needed for a score.
All types combined: 79 (likely 74 to 84, high confidence), from 102 claims: 64 factual claims, 38 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Gary Marcus's claims
52 claims
““Alignment” between what humans want and what machines do will continue to be a critical, unsolved problem.”
Gary Marcus, author 25 December 2022 What to Expect When You're Expecting … GPT-4 (Marcus on AI)
AI alignment between human wants and machine behavior will remain a critical unsolved problem, with GPT-4 unable to reliably follow shared human values.
Mostly true True in substance, wrong in a detail.
What would count as true or false
Judge against GPT-4 as released (March 2023) and the state of the field during 2023. True: through 2023 alignment remained widely recognised as an unsolved, critical problem, and GPT-4 demonstrably failed to reliably follow human values and intentions, e.g. jailbreaks, harmful outputs, documented misbehaviour in OpenAI's system card or independent red-teaming. Mostly true: both parts hold, but GPT-4's failures were notably less frequent than the claim implies (e.g. it mostly behaved well, with failures mainly under adversarial prompting). Mixed: one part holds and the other does not, e.g. alignment remained unsolved in general but GPT-4 was shown to follow shared human values reliably in practice. Mostly false: GPT-4 was broadly reliable in following human values with only rare, narrow failures, and experts considered practical alignment largely achieved for such models. False: alignment was widely regarded as solved and GPT-4 reliably followed human values.
Evidence
- GPT-4 System Card - cdn.openai.com OpenAI
“Training on this data improves model robustness but does not fully solve the problem of “jailbreaks” leading to harmful content.”
- GPT-4 System Card - cdn.openai.com OpenAI
“In most cases, GPT-4-launch exhibits much safer behavior due to the safety mitigations we applied.”
- Model alignment protects against accidental harms, not intentional ones normaltech.ai (Arvind Narayanan, Sayash Kapoor, Seth Lazar) · 1 December 2023
“RLHF has very substantially solved this problem, and by solving it OpenAI created a multi-billion dollar industry where one didn't exist before.”
- Model alignment protects against accidental harms, not intentional ones normaltech.ai (Arvind Narayanan, Sayash Kapoor, Seth Lazar) · 1 December 2023
“A wave of recent research has shown that RLHF is brittle.”
Reasoning
OpenAI's GPT-4 System Card says its safety training "does not fully solve the problem of 'jailbreaks' leading to harmful content" (E1), and later independent analysis calls RLHF "brittle" (E4). Both support the claim that alignment stayed an unsolved problem. However, the same system card says the launch version of GPT-4 showed "much safer behavior" in most cases (E2), and Narayanan, Kapoor and Lazar write that RLHF "very substantially solved" the problem of accidental harms (E3). The failures therefore appear mainly under adversarial prompting rather than in ordinary use, which matches the criteria's description of "mostly true".
Cite this claim
Gary Marcus, “AI alignment between human wants and machine behavior will remain a critical unsolved problem, with GPT-4 unable to reliably follow shared human values.” (said 25 December 2022). Verdict: mostly true. Reliability API, claim c_d30cb5d57cfa, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_d30cb5d57cfa
Claims without a verdict
Is a verdict here wrong, or do you speak for Gary Marcus? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.