Researchers test 241 large language models on existential risk estimates
Researchers tested 241 large language models on their estimates of existential risk from advanced artificial intelligence, finding a median estimate of 10.0% with the central 90% of responses ranging from 3.5% to 15.8%.
More capable models tended to estimate higher existential risks, while newer models showed more consistent estimates without necessarily higher risk assessments.
Loss of control over artificial intelligence was the cause most frequently cited by the tested models as the likely source of existential risk.
The researchers note that the models' responses likely reflect their training data, including published expert surveys, and the choices made in aligning the systems, rather than independent judgments.
Sources behind this report
Report history
No earlier revision dates are retained here. This does not establish that the report has never changed; read any dated updates and corrections.
This brief was made by AI and checked against its cited public sources. How the newsroom works