🔔 New paper from OpenAI - this is big for AI.

If we can understand why AI hallucinates, we’ll be able to build models that don’t hallucinate at all. Large language models often guess when uncertain, producing convincing but false answers. This happens because current training and evaluation reward confidence over honesty.

Hallucinations stem from statistical errors, reinforced by benchmarks that penalize admitting uncertainty. To fix this, the paper authors propose changing how benchmarks score uncertainty, which could steer the field toward more trustworthy AI.

@aipost 🪙 | Our X 🥇

Geoffrey Hinton: ‘AI will make a few people much richer and most people poorer’

No lie detected.

@aipost 🪙 | Our X 🥇

Professor Roman Yampolskiy predicts AGI by 2027.

@aipost 🪙 | Our X 🥇

Join: @davidavocadowolfe 🥑