Here is what the page “Claude Opus 4.8: Can We Trust AI Not to Lie?” contains. Anthropic's latest AI model, Claude Opus 4.8, claims to be more honest and less prone to deception than its predecessors. In this video, we examine the evidence: how the model was trained using constitutional AI and reinforcement learning from human feedback to reduce sycophancy and factual errors. We also discuss the benchmarks used to measure truthfulness, including the TruthfulQA dataset. But is it truly a 'lying machine no more'?
We analyze cases where the model still exhibits subtle biases or confabulations, and what that means for real-world deployment. The video also covers the broader implications for AI safety and the ongoing challenge of aligning language models with human values. Sponsored by Lambda, we also touch on the computational resources needed to train such models. This listing links straight to the original page, where the content is available in full.